Ad Hoc Networks 3 (2005) 257–279 www.elsevier.com/locate/adhoc

Underwater acoustic sensor networks: research challenges

Ian F. Akyildiz *, Dario Pompili, Tommaso Melodia

Broadband and Wireless Networking Laboratory, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA

Received 15 July 2004; received in revised form 20 September 2004; accepted 21 January 2005 Available online 2 February 2005

Abstract

Underwater sensor nodes will find applications in oceanographic data collection, pollution monitoring, offshore exploration, disaster prevention, assisted navigation and tactical surveillance applications. Moreover, unmanned or autonomous underwater vehicles (UUVs, AUVs), equipped with sensors, will enable the exploration of natural under- sea resources and gathering of scientific data in collaborative monitoring missions. Underwater acoustic networking is the enabling technology for these applications. Underwater networks consist of a variable number of sensors and vehi- cles that are deployed to perform collaborative monitoring tasks over a given area. In this paper, several fundamental key aspects of underwater acoustic communications are investigated. Different archi- tectures for two-dimensional and three-dimensional underwater sensor networks are discussed, and the characteristics of the underwater channel are detailed. The main challenges for the development of efficient networking solutions posed by the underwater environment are detailed and a cross-layer approach to the integration of all communication functionalities is suggested. Furthermore, open research issues are discussed and possible solution approaches are outlined. 2005 Published by Elsevier B.V.

Keywords: Underwater acoustic sensor networks; Underwater networking; Acoustic communications

1. Introduction lection, pollution monitoring, offshore explora- tion, disaster prevention, assisted navigation Underwater sensor networks are envisioned to and tactical surveillance applications. Multiple enable applications for oceanographic data col- unmanned or autonomous underwater vehicles (UUVs, AUVs), equipped with underwater sen- sors, will also find application in exploration * Corresponding author. Tel.: +1 404 894 5141; fax: +1 404 of natural undersea resources and gathering of 894 7883. E-mail addresses: [email protected] (I.F. Akyildiz), dar- scientific data in collaborative monitoring mis- [email protected] (D. Pompili), [email protected] (T. sions. To make these applications viable, there Melodia). is a need to enable underwater communications

1570-8705/$ - see front matter 2005 Published by Elsevier B.V. doi:10.1016/j.adhoc.2005.01.004 258 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 among underwater devices. Underwater sensor ture gradients (thermoclines), which are consid- nodes and vehicles must possess self-configura- ered to be a breeding ground for certain marine tion capabilities, i.e., they must be able to micro-organisms. coordinate their operation by exchanging config- • Undersea explorations. Underwater sensor net- uration, location and movement information, works can help detecting underwater oilfields and to relay monitored data to an onshore or reservoirs, determine routes for laying under- station. sea cables, and assist in exploration for valuable Wireless underwater acoustic networking is the minerals. enabling technology for these applications. Under- • Disaster prevention. Sensor networks that mea- Water Acoustic Sensor Networks (UW-ASNs) sure seismic activity from remote locations can consist of a variable number of sensors and provide tsunami warnings to coastal areas [42], vehicles that are deployed to perform collaborative or study the effects of submarine earthquakes monitoring tasks over a given area. To achieve (seaquakes). this objective, sensors and vehicles self-organize • Assisted navigation. Sensors can be used to iden- in an autonomous network which can tify hazards on the seabed, locate dangerous adapt to the characteristics of the ocean environ- rocks or shoals in shallow waters, mooring posi- ment [1]. tions, submerged wrecks, and to perform The above described features enable a broad bathymetry profiling. range of applications for underwater acoustic sen- • Distributed tactical surveillance. AUVs and sor networks: fixed underwater sensors can collaboratively monitor areas for surveillance, reconnaissance, • Ocean sampling networks. Networks of sensors targeting and intrusion detection systems. For and AUVs, such as the Odyssey-class AUVs example, in [15], a 3D underwater sensor net- [2], can perform synoptic, cooperative adaptive work is designed for a tactical surveillance sampling of the 3D coastal ocean environment system that is able to detect and classify subma- [3]. Experiments such as the Monterey Bay rines, small delivery vehicles (SDVs) and divers field experiment [4] demonstrated the advanta- based on the sensed data from mechanical, ges of bringing together sophisticated new radiation, magnetic and acoustic microsensors. robotic vehicles with advanced ocean models With respect to traditional radar/sonar sys- to improve the ability to observe and pre- tems, underwater sensor networks can reach a dict the characteristics of the oceanic envi- higher accuracy, and enable detection and ronment. classification of low signature targets by also • Environmental monitoring. UW-ASNs can per- combining measures from different types of form pollution monitoring (chemical, biologi- sensors. cal and nuclear). For example, it may be • Mine reconnaissance. The simultaneous opera- possible to detail the chemical slurry of antibi- tion of multiple AUVs with acoustic and opti- otics, estrogen-type hormones and insecticides cal sensors can be used to perform rapid to monitor streams, rivers, lakes and ocean environmental assessment and detect mine-like bays (water quality in situ analysis) [51]. Moni- objects. toring of ocean currents and winds, improved weather forecast, detecting climate change, Underwater networking is a rather unexplored under-standing and predicting the effect of area although underwater communications have human activities on marine ecosystems, biolog- been experimented since World War II, when, in ical monitoring such as tracking of fishes or 1945, an underwater telephone was developed in micro-organisms, are other possible applica- the United States to communicate with submarines tions. For example, in [52], the design and con- [39]. Acoustic communications are the typical struction of a simple underwater sensor physical layer technology in underwater networks. network is described to detect extreme tempera- In fact, radio waves propagate at long distances I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 259 through conductive sea water only at extra low fre- selected ocean areas, remote configuration and quencies (30–300 Hz), which require large anten- interaction with onshore human operators. This nae and high transmission power. For example, can be obtained by connecting underwater instru- the Berkeley Mica 2 Motes, the most popular ments by means of wireless links based on acoustic experimental platform in the sensor networking communication. community, have been reported to have a trans- Many researchers are currently engaged in mission range of 120 cm in underwater at developing networking solutions for terrestrial 433 MHz by experiments performed at the Ro- wireless ad hoc and sensor networks. Although botic Embedded Systems Laboratory (RESL) at there exist many recently developed network pro- the University of Southern California. Optical tocols for wireless sensor networks, the unique waves do not suffer from such high attenuation characteristics of the underwater acoustic commu- but are affected by scattering. Moreover, transmis- nication channel, such as limited bandwidth sion of optical signals requires high precision in capacity and variable delays [38], require very effi- pointing the narrow laser beams. Thus, links in cient and reliable new data communication underwater networks are based on acoustic wire- protocols. less communications [45]. Major challenges in the design of underwater The traditional approach for ocean-bottom or acoustic networks are: ocean-column monitoring is to deploy underwater sensors that record data during the monitoring • The available bandwidth is severely limited; mission, and then recover the instruments [37]. • The underwater channel is severely im- This approach has the following disadvantages: paired, especially due to multi-path and fad- ing; • No real-time monitoring. The recorded data can- • Propagation delay in underwater is five orders not be accessed until the instruments are recov- of magnitude higher than in radio frequency ered, which may happen several months after (RF) terrestrial channels, and extremely the beginning of the monitoring mission. This variable; is critical especially in surveillance or in envi- • High bit error rates and temporary losses of ronmental monitoring applications such as seis- connectivity (shadow zones) can be experienced, mic monitoring. due to the extreme characteristics of the under- • No on-line system reconfiguration. Interaction water channel; between onshore control systems and the mon- • Battery power is limited and usually batteries itoring instruments is not possible. This cannot be recharged, also because solar energy impedes any adaptive tuning of the instruments, cannot be exploited; nor is it possible to reconfigure the system after • Underwater sensors are prone to failures particular events occur. because of fouling and corrosion. • No failure detection. If failures or misconfigura- tions occur, it may not be possible to detect In this survey, we discuss several fundamental them before the instruments are recovered. This key aspects of underwater acoustic communica- can easily lead to the complete failure of a mon- tions. We discuss the communication architecture itoring mission. of underwater sensor networks as well as the fac- • Limited storage capacity. The amount of data tors that influence underwater network design. that can be recorded during the monitoring mis- The ultimate objective of this paper is to encour- sion by every sensor is limited by the capacity of age research efforts to lay down fundamental basis the onboard storage devices (memories, hard for the development of new advanced communica- disks). tion techniques for efficient underwater communi- cation and networking for enhanced ocean Therefore, there is a need to deploy underwater monitoring and exploration applications. In networks that will enable real-time monitoring of Table 3, we report a list of research laboratories 260 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 and ongoing research projects related to underwa- organize the network topology such a way that ter communications and explorations. no communication bottleneck is introduced. The remainder of this paper is organized as The communication architectures introduced follows. In Sections 2 and 3 we introduce the com- here are used as a basis for discussion of the chal- munication architecture and design challenges, lenges associated with underwater acoustic sensor respectively, of underwater acoustic sensor networks. The underwater sensor network topol- networks. In Section 4, we investigate the under- ogy is an open research issue in itself that needs water acoustic communication channel and further analytical and simulative investigation summarize the associated physical layer challenges from the research community. In the remainder for underwater networking. In Sections 5–9 we dis- of this section, we discuss the following cuss physical, data link, network, transport and architectures: application layer issues in underwater sensor net- works, respectively. In Section 10 we describe • Static two-dimensional UW-ASNs for ocean bot- some experimental implementations of underwater tom monitoring. These are constituted by sensor sensor networks while in Section 11 we draw the nodes that are anchored to the bottom of the main conclusions. ocean, as discussed in Section 2.1. Typical applications may be environmental monitoring, or monitoring of underwater plates in tectonics 2. Underwater acoustic sensor networks: [21]. communication architecture • Static three-dimensional UW-ASNs for ocean- column monitoring. These include networks of In this section, we describe the communication sensors whose depth can be controlled by means architecture of underwater acoustic sensor of techniques discussed in Section 2.2, and may networks. In particular, we introduce reference be used for surveillance applications or moni- architectures for two-dimensional and three- toring of ocean phenomena (ocean bio–geo- dimensional underwater networks, and present chemical processes, water streams, pollution). several types of autonomous underwater vehicles • Three-dimensional networks of autonomous (AUVs) which can enhance the capabilities of underwater vehicles (AUVs). These networks underwater sensor networks. include fixed portions composed of anchored The network topology is in general a crucial sensors and mobile portions constituted by factor in determining the energy consumption, the autonomous vehicles, as detailed in Section 2.3. capacity and the reliability of a network. Hence, the network topology should be carefully engi- neered and post-deployment topology optimization 2.1. Two-dimensional underwater sensor networks should be performed, when possible. Underwater monitoring missions can be extre- A reference architecture for two-dimensional mely expensive due to the high cost of underwater underwater networks is shown in Fig. 1. A group devices. Hence, it is important that the deployed of sensor nodes are anchored to the bottom of network be highly reliable, so as to avoid failure the ocean with deep ocean anchors. Underwater of monitoring missions due to failure of single or sensor nodes are interconnected to one or more multiple devices. For example, it is crucial to avoid underwater sinks (uw-sinks) by means of wireless designing the network topology with single points acoustic links. Uw-sinks, as shown in Fig. 1, are of failure that could compromise the overall func- network devices in charge of relaying data from tioning of the network. the ocean bottom network to a surface station. The network capacity is also influenced by the To achieve this objective, uw-sinks are equipped network topology. Since the capacity of the under- with two acoustic transceivers, namely a vertical water channel is severely limited, as will be dis- and a horizontal transceiver. The horizontal trans- cussed in Section 4, it is very important to ceiver is used by the uw-sink to communicate with I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 261

Fig. 1. Architecture for 2D underwater sensor networks. the sensor nodes in order to: (i) send commands work throughput because of increased acoustic and configuration data to the sensors (uw-sink to interference due to high transmission power. In sensors); (ii) collect monitored data (sensors to case of multi-hop paths, as in terrestrial sensor uw-sink). The vertical link is used by the uw-sinks networks [10], the data produced by a source sen- to relay data to a surface station. In deep water sor is relayed by intermediate sensors until it applications, vertical transceivers must be long reaches the uw-sink. This may result in energy range transceivers as the ocean can be as deep as savings and increased network capacity, but in- 10 km. The surface station is equipped with an creases the complexity of the routing functional- acoustic transceiver that is able to handle multiple ity. In fact, every network device usually takes parallel communications with the deployed uw- part in a collaborative process whose objective is sinks. It is also endowed with a long range RF to diffuse topology information such that efficient and/or satellite transmitter to communicate with and loop free routing decisions can be made at the onshore sink (os-sink) and/or to a surface sink each intermediate node. This process involves sig- (s-sink). naling and computation. Since energy and capacity Sensors can be connected to uw-sinks via direct are precious resources in underwater environments, links or through multi-hop paths. In the former as discussed above, in UW-ASNs the objective is case, each sensor directly sends the gathered data to deliver event features by exploiting multi-hop to the selected uw-sink. However, in UW-ASNs, paths and minimizing the signaling overhead neces- the power necessary to transmit may decay with sary to construct underwater paths at the same powers greater than two of the distance [44],and time. the uw-sink may be far from the sensor node. Consequently, although direct link connection is 2.2. Three-dimensional underwater sensor networks the simplest way to network sensors, it may not be the most energy efficient solution. Further- Three dimensional underwater networks are more, direct links are very likely to reduce the net- used to detect and observe phenomena that cannot 262 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 be adequately observed by means of ocean bottom that connects the sensor to the anchor, by means sensor nodes, i.e., to perform cooperative sampling of an electronically controlled engine that resides of the 3D ocean environment. In three-dimen- on the sensor. A challenge to be addressed in such sional underwater networks, sensor nodes float at an architecture is the effect of ocean currents on different depths in order to observe a given phe- the described mechanism to regulate the depth of nomenon. One possible solution would be to the sensors. attach each uw-sensor node to a surface buoy, by Many challenges arise with such an architec- means of wires whose length can be regulated so ture, that need to be solved in order to enable as to adjust the depth of each sensor node [15]. 3D monitoring, including: However, although this solution allows easy and quick deployment of the sensor network, multiple • Sensing coverage. Sensors should collabora- floating buoys may obstruct ships navigating on tively regulate their depth in order to achieve the surface, or they can be easily detected and 3D coverage of the ocean column, according deactivated by enemies in military settings. Fur- to their sensing ranges. Hence, it must be possi- thermore, floating buoys are vulnerable to weather ble to obtain sampling of the desired phenome- and tampering or pilfering. non at all depths. For these reasons, a different approach can be • Communication coverage. Since in 3D underwa- to anchor sensor devices to the bottom of the ter networks there may be no notion of uw-sink, ocean. In this architecture, depicted in Fig. 2, each sensors should be able to relay information to sensor is anchored to the ocean bottom and the surface station via multi-hop paths. Thus, equipped with a floating buoy that can be inflated network devices should coordinate their depths by a pump. The buoy pushes the sensor towards in such a way that the network topology is the ocean surface. The depth of the sensor can then always connected, i.e., at least one path from be regulated by adjusting the length of the wire every sensor to the surface station always exists.

Fig. 2. Architecture for 3D underwater sensor networks. I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 263

Sensing and communication coverage in a 3D One of the design objectives of AUVs is to environment are rigorously investigated in [40]. make them rely on local intelligence and less The diameter, minimum and maximum degree of dependent on communications from online shores the reachability graph that describes the network [25]. In general, control strategies are needed for are derived as a function of the communication autonomous coordination, obstacle avoidance range, while different degrees of coverage for the and steering strategies. Solar energy systems allow 3D environment are characterized as a function of increasing the lifetime of AUVs, i.e., it is not nec- the sensing range. These techniques could be essary to recover and recharge the vehicle on a dai- exploited to investigate the coverage issues in ly basis. Hence, solar powered AUVs can acquire UW-ASNs. continuous information for periods of time of the order of months [27]. 2.3. Sensor networks with autonomous Several types of AUVs exist as experimental underwater vehicles platforms for underwater experiments. Some of them resemble small-scale submarines (such as AUVs can function without tethers, cables, or the Odyssey-class AUVs [2] developed at MIT). remote control, and therefore they have a multi- Others are simpler devices that do not encompass tude of applications in oceanography, environ- such sophisticated capabilities. For example, drift- mental monitoring, and underwater resource ers and gliders are oceanographic instruments of- study. Previous experimental work has shown the ten used in underwater explorations. Drifter feasibility of relatively inexpensive AUV subma- underwater vehicles drift with local current and rines equipped with multiple underwater sensors have the ability to move vertically through the that can reach any depth in the ocean [2]. Hence, water column. They are used for taking measure- they can be used to enhance the capabilities of ments at preset depths [24]. Underwater gliders underwater sensor networks in many ways. The [18] are battery powered autonomous underwater integration and enhancement of fixed sensor net- vehicles that use hydraulic pumps to vary their vol- works with AUVs is an almost unexplored re- ume by a few hundred cubic centimeters in order search area which requires new network to generate the buoyancy changes that power their coordination algorithms such as: forward gliding. When they emerge on the surface, global positioning system (GPS) is used to locate • Adaptive sampling. This includes control strate- the vehicle. This information can be relayed to gies to command the mobile vehicles to places the onshore station while operators can interact where their data will be most useful. This by sending control information to the gliders. approach is also known as adaptive sampling Depth capabilities range from 200 m to 1500 m and has been proposed in pioneering monitoring while operating lifetimes range from a few weeks missions such as [4]. For example, the density to several months. These long durations are possi- of sensor nodes can be adaptively increased ble because gliders move very slowly, typically in a given area when a higher sampling rate is 25 cm/s (0.5 knots). In [34], a control strategy for needed for a given monitored phenomenon. groups of gliders to cooperatively move and recon- • Self-configuration. This includes control proce- figure in response to a sensed distributed environ- dures to automatically detect connectivity holes ment is presented. The proposed framework allows due to node failures or channel impairment and preserving the symmetry of the group of gliders. request the intervention of an AUV. Further- The group is constrained to maintain a uniform more, AUVs can either be used for installation distribution as needed, but is free to spin and pos- and maintenance of the sensor network infra- sibly wiggle with the current. In [20], results are re- structure or to deploy new sensors. They can ported on the application of the theory in [34] on a also be used as temporary relay nodes to restore fleet of autonomous underwater gliders during the connectivity. experiment on Monterey Bay in 2003 [4]. 264 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279

3. Underwater acoustic sensor networks: design challenges

In this section, we describe the design challenges of underwater acoustic sensor networks. In partic- ular, we itemize the main differences between terres- trial and underwater sensor networks, we detail key design issues and deployment challenges for under- water sensors, and we give motivations for a cross- layer design approach to improve the network efficiency in the critical underwater environment.

3.1. Differences with terrestrial sensor networks

The main differences between terrestrial and underwater sensor networks are as follows: Fig. 3. Internal architecture of an underwater sensor node.

• Cost. While terrestrial sensor nodes are controller/CPU which is interfaced with an ocean- expected to become increasingly inexpensive, ographic instrument or sensor through a sensor underwater sensors are expensive devices. This interface circuitry. The controller receives data is especially due to the more complex underwa- from the sensor and it can store it in the onboard ter transceivers and to the hardware protection memory, process it, and send it to other network needed in the extreme underwater environment. devices by controlling the acoustic modem. The • Deployment. While terrestrial sensor networks electronics are usually mounted on a frame which are densely deployed, in underwater, the is protected by a PVC housing. Sometimes all sen- deployment is deemed to be more sparse, due sor components are protected by bottom-mounted to the cost involved and to the challenges asso- instrument frames that are designed to permit azi- ciated to the deployment itself. muthally omnidirectional acoustic communica- • Power. The power needed for acoustic under- tions, and protect sensors and modems from water communications is higher than in terres- potential impact of trawling gear, especially in trial radio communications due to higher areas subjected to fishing activities. In [16], the distances and to more complex signal process- protecting frame is designed so as to deflect trawl- ing at the receivers to compensate for the ing gear on impact, by housing all components impairments of the channel. beneath a low-profile pyramidal frame. • Memory. While terrestrial sensor nodes have Underwater sensors include sensors to measure very limited storage capacity, uw-sensors may the quality of water and to study its characteristics need to be able to do some data caching as such as temperature, density, salinity (interfero- the underwater channel may be intermittent. metric and refractometric sensors), acidity, chemi- • Spatial correlation. While the readings from ter- cals, conductivity, pH (magnetoelastic sensors), restrial sensors are often correlated, this is more oxygen (Clark-type electrode), hydrogen, dissolved unlikely to happen in underwater networks due methane gas (METS), and turbidity. Disposable to the higher distance among sensors. sensors exist that detect ricin, the highly poisonous protein found in castor beans and thought to be a potential terrorism agent. DNA microarrays can 3.2. Underwater sensors be used to monitor both abundance and activity level variations among natural microbial popula- The typical internal architecture of an underwa- tions. Other existing underwater sensors include ter sensor is shown in Fig. 3. It consists of a main hydrothermal sulfide, silicate, voltammetric sensors I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 265 for spectrophotometry, gold-amalgam electrode layer functionalities. The protocol stack should sensors for sediment measurements of metal also include a power management plane,acoordina- ions (ion-selective analysis), amperometric micro- tion plane, and a localization plane. The power sensors for H2S measurements for studies of management plane is responsible for network anoxygenic photosynthesis, sulfide oxidation, and functionalities aimed at minimizing the energy sulfate reduction of sediments. In addition, force/ consumption (e.g., sleep modes, power control, torque sensors for underwater applications requir- etc.). The coordination plane is responsible for ing simultaneous measurements of several forces all functionalities that require coordination among and moments have also been developed, as well sensors (e.g., coordination of the sleep modes, data as quantum sensors to measure light radiation aggregation, 3D topology optimization). The and sensors for measurements of harmful algal localization plane is responsible for providing blooms. absolute or relative localization information to The challenges related to the deployment of low the sensor node, when needed by the protocol cost, low scale underwater sensors, are listed as stack or by the application. follows: While all the research on underwater network- ing so far has followed the traditional layered ap- • It is necessary to develop less expensive, robust proach for network design, it is an increasingly ‘‘nano-sensors’’, e.g., sensors based on nano- accepted opinion in the wireless networking com- technology, which involves development of munity that the improved network efficiency, espe- materials and systems at the atomic, molecular, cially in critical environments, can be obtained with or macromolecular levels in the dimension a cross-layer design approach. These techniques range of approximately 1–500 nm. will entail a joint design of different network func- • It is necessary to devise periodical cleaning tionalities, from modem design to MAC and rout- mechanisms against corrosion and fouling, ing, from channel coding and modulation to source which may impact the lifetime of underwater compression and transport layer, with the objective devices. For example, some sensors for pCO2, to overcome the shortcomings of a layered ap- pH and nitrate measurement, and fluorometers proach that lacks of information sharing across and spectral radiometers, may be limited by protocol layers, forcing the network to operate in bio-fouling, especially on a long time scale. a suboptimal mode. Hence, while in the following • There is a need for robust, stable sensors on a sections for the sake of clarity we present the chal- high range of temperatures since sensor drift lenges associated with underwater sensor networks of underwater devices may be a concern. To this following the traditional layered approach, we be- end, protocols for in situ calibration of sensors lieve that the underwater environment particularly to improve accuracy and precision of sampled requires for cross-layer design solutions that allow data must be developed. a more efficient use of the scarce available • There is a need for new integrated sensors for resources. However, although we advocate inte- synoptic sampling of physical, chemical, and grating functionalities to improve network perfor- biological parameters to improve the under- mance and to avoid duplication of functions by standing of processes in marine systems. means of cross-layer design, it is important to con- sider the ease of design by following a modular design approach. This also allows improving and 3.3. A Cross-layer protocol stack upgrading particular functionalities without the need to re-design the entire communication system. A protocol stack for uw-sensors should com- Although systematic research on cross-layer de- bine power awareness and management, and pro- sign for underwater communications is missing, a mote cooperation among the sensor nodes. It study on the interaction between physical and should consist of physical layer, data link layer, MAC layers is presented in [29], where a method network layer, transport layer, and application is proposed based on the sonar equation [49] to 266 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279

90

80

70

60

50

40

30 Acoustic Power Attenuation [dB] 20 50 40 1000 30 800 20 600 400 10 200 0 0 Frequency [KHz] Distance [m]

Fig. 4. Path loss of short range shallow UW-A channels vs distance and frequency in band 1–50 kHz. estimate the battery lifetime and power cost ity when it is lost and that react to unpaired or for shallow water underwater acoustic sensor congested links by taking appropriate action networks for civilian applications. The battery life- (e.g., dynamical rerouting) in order to meet the time is modeled as dependent on four key parame- given delay bound. Conversely, other applications ters, namely internode distance, transmission may produce large bundles of data to be delivered frequency, frequency of data updates and number to the onshore sink without particular delay con- of nodes per cluster. Interestingly, since in shallow straints. With this respect, the Delay-Tolerant water the acoustic propagation loss increases with Networking Research Group (DTNRG) [5,19] increasing frequency and distance (as shown in developed mechanisms to resolve the intermittent Fig. 4), it is proposed to assign lower frequencies connectivity, long or variable delay, asymmetric to sensor nodes that are closer to the sink, since data rates, and high error rates by using a store they also have to relay data on behalf of more dis- and forward mechanism based on a middleware tant nodes. This way, the energy consumption is between the application layer and the lower layers. somehow equalized and the network lifetime is Similar methodologies may be particularly useful prolonged. for applications, such as those that record seismic activity, that have a very low duty cycle and pro- 3.4. Real-time vs delay-tolerant networking duce, when activated, large bundles of data that need to be relayed to a monitoring station where As in terrestrial sensor networks, depending on it can be analyzed to predict future activity. On the application, there may be very different the other hand, sensor networks intended for requirements for data delivery. For example, sur- disaster prevention such as those that provide veillance application may need very fast reaction earthquake or tsunami warnings, require immedi- to events and thus networking protocols that ate delivery of information and hence real-time provide guaranteed delay-bounded delivery are re- protocols. Therefore, the design of networking quired. Hence, it is necessary to develop protocols solutions for underwater acoustic sensor networks that deal with the characteristics of the underwater should always be aware of the difference between environment in order to quickly restore connectiv- real-time and delay-tolerant applications, and I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 267 jointly tune existing solutions to the application challenges posed by the underwater channel for needs and to the characteristics of the underwater underwater sensor networking. These include: environment. • Path loss:  Attenuation. Is mainly provoked by absorp- 4. Basics of acoustic propagation tion due to conversion of acoustic energy into heat. The attenuation increases with distance Underwater acoustic communications are and frequency. Fig. 4 shows the acoustic mainly influenced by path loss, noise, multi-path, attenuation with varying frequency and dis- Doppler spread, and high and variable propagation tance for a short range shallow water UW-A delay. All these factors determine the temporal channel, according to the propagation model and spatial variability of the acoustic channel, in [49]. The attenuation is also caused by scat- and make the available bandwidth of the Under- tering and reverberation (on rough ocean sur- Water Acoustic channel (UW-A) limited and face and bottom), refraction, and dispersion dramatically dependent on both range and fre- (due to the displacement of the reflection quency. Long-range systems that operate over sev- point caused by wind on the surface). Water eral tens of kilometers may have a bandwidth of depth plays a key role in determining the only a few kHz, while a short-range system operat- attenuation. ing over several tens of meters may have more than  Geometric spreading. This refers to the a hundred kHz of bandwidth. In both cases these spreading of sound energy as a result of the factors lead to low bit rate [14], in the order of tens expansion of the wavefronts. It increases with of kbit/s for existing devices. the propagation distance and is independent Underwater acoustic communication links can of frequency. There are two common kinds be classified according to their range as very long, of geometric spreading: spherical (omni-direc- long, medium, short, and very short links [45]. Table tional point source), which characterizes deep 1 shows typical bandwidths of the underwater water communications, and cylindrical (hori- channel for different ranges. Acoustic links are also zontal radiation only), which characterizes roughly classified as vertical and horizontal, shallow water communications. according to the direction of the sound ray with re- • Noise: spect to the ocean bottom. As will be shown later,  Man made noise. This is mainly caused by their propagation characteristics differ consider- machinery noise (pumps, reduction gears, ably, especially with respect to time dispersion, power plants), and shipping activity (hull multi-path spreads, and delay variance. In the fol- fouling, animal life on hull, cavitation), espe- lowing, as usually done in oceanic literature, shal- cially in areas encumbered with heavy vessel low water refers to water with depth lower than traffic. 100 m, while deep water is used for deeper oceans.  Ambient noise. Is related to hydrodynamics Hereafter we analyze the factors that influence (movement of water including tides, current, acoustic communications in order to state the storms, wind, and rain), and to seismic and biological phenomena. In [23], boat noise and snapping shrimps have been found to Table 1 be the primary sources of noise in shallow Available bandwidth for different ranges in UW-A channels water by means of measurement experiments Range [km] Bandwidth [kHz] on the ocean bottom. Very long 1000 <1 • Multi-path: Long 10–100 2–5  Multi-path propagation may be responsible Medium 1–10 10 for severe degradation of the acoustic com- Short 0.1–1 20–50 munication signal, since it generates inter- Very short <0.1 >100 symbol interference (ISI). 268 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279

 The multi-path geometry depends on the link together with the wave guide nature of the chan- configuration. Vertical channels are charac- nel, cause the acoustic channel to be highly tempo- terized by little time dispersion, whereas hor- rally and spatially variable. In particular, the izontal channels may have extremely long horizontal channel is by far more rapidly varying multi-path spreads. than the vertical channel, in both deep and shallow  The extent of the spreading is a strong func- water. tion of depth and the distance between trans- mitter and receiver. • High delay and delay variance: 5. Physical layer  The propagation speed in the UW-A channel is five orders of magnitude lower than in the Until the beginning of the last decade, due to radio channel. This large propagation delay the challenging characteristics of the underwater (0.67 s/km) can reduce the throughput of the channel, underwater modem development was system considerably. based on non-coherent frequency shift keying  The very high delay variance is even more (FSK) modulation, since it relies on energy detec- harmful for efficient protocol design, as it pre- tion. Thus, it does not require phase tracking, vents from accurately estimating the round which is a very difficult task mainly because of trip time (RTT), which is the key parameter the Doppler-spread in the UW-A channel, de- for many common communication protocols. scribed in Section 4. In FSK modulation schemes • Doppler spread: developed for underwater communications, the  The Doppler frequency spread can be signif- multi-path effects are suppressed by inserting time icant in UW-A channels [45], thus causing a guards between successive pulses to ensure that the degradation in the performance of digital reverberation, caused by the rough ocean surface communications: high data rate transmis- and bottom, vanishes before each subsequent sions cause adjacent symbols to interfere pulse is received. Dynamic frequency guards can at the receiver. This requires sophisticated also be used between frequency tones to adapt signal processing to deal with the generated the communication to the Doppler spreading of ISI. the channel. Although non-coherent modulation  The Doppler spreading generates a simple fre- schemes are characterized by a high power effi- quency translation, which is relatively easy ciency, their low bandwidth efficiency makes them for a receiver to compensate for; and a contin- unsuitable for high data rate multiuser networks. uous spreading of frequencies that constitutes Hence, coherent modulation techniques have been a non-shifted signal, which is more difficult to developed for long-range, high-throughput sys- compensate for. tems. In the last years, fully coherent modulation  If a channel has a Doppler spread with techniques, such as phase shift keying (PSK) and bandwidth B and a signal has symbol dura- quadrature amplitude modulation (QAM), have tion T, then there are approximately BT become practical due to the availability of power- uncorrelated samples of its complex enve- ful digital processing. Channel equalization tech- lope. When BT is much less than unity, the niques are exploited to leverage the effect of the channel is said to be underspread and the inter-symbol interference (ISI), instead of trying effects of the Doppler fading can be ignored, to avoid or suppress it. Decision-feedback equaliz- while, if greater than unity, it is said to be ers (DFEs) track the complex, relatively slowly overspread [32]. varying channel response and thus provide high throughput when the channel is slowly varying. Most of the described factors are caused by the Conversely, when the channel varies faster, it is chemical-physical properties of the water medium necessary to combine the DFE with a Phase such as temperature, salinity and density, and by Locked Loop (PLL) [46], which estimates and their spatio-temporal variations. These variations, compensates for the phase offset in a rapid, stable I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 269

Table 2 non-coherent modems to the recent coherent Evolution of modulation technique modems. Type Year Rate [kbps] Band [kHz] Range [km]a Differential phase shift keying (DPSK) serves as

FSK 1984 1.2 5 3s an intermediate solution between incoherent and PSK 1989 500 125 0.06d fully coherent systems in terms of bandwidth effi- FSK 1991 1.25 10 2d ciency. DPSK encodes information relative to the PSK 1993 0.3–0.5 0.3–1 200d–90s previous symbol rather than to an arbitrary fixed PSK 1994 0.02 20 0.9 s reference in the signal phase and may be referred FSK 1997 0.6–2.4 5 10d–5s DPSK 1997 20 10 1d to as a partially coherent modulation. While this PSK 1998 1.67–6.7 2–10 4d–2s strategy substantially alleviates carrier phase-track- 16-QAM 2001 40 10 0.3s ing requirements, the penalty is an increased error a The subscripts d and s stand for deep and shallow water. probability over PSK at an equivalent data rate. With respect to Table 2, it is worth noticing that early phase-coherent systems achieved higher manner. The use of decision feedback equalization bandwidth efficiencies (bit rate/occupied band- and phase-locked loops is driven by the complexity width) than their incoherent counterparts, but and time variability of ocean channel impulse re- they did not outperform incoherent modulation sponses. Table 2 presents the evolution from schemes yet. In fact, coherent systems had lower

Table 3 Research laboratories and ongoing research projects related to underwater acoustic sensor networks Research lab or project name Research area URL BWN-Lab @ GeorgiaTech Underwater acoustic sensor networks http://www.ece.gatech.edu/research/labs/ bwn/UWASN/ MIT & Woods Hole O.I. Underwater acoustic networks http://www.mit.edu/people/millitsa/ research.html Front Project @ UConn Spatial sampling of ocean http://www.nopp.uconn.edu/ADCP/ index.html Autonomous Ocean Sampling Networks II Adaptive ocean sampling http://www.princeton.edu/dcsl/aosn/ Adaptive Sampling and Prediction (ASAP) Adaptive ocean sampling http://www.princeton.edu/dcsl/asap/ Sensor Networks for Undersea Seismic Underwater acoustic sensor networks http://www.isi.edu/ilense/snuse/ Experimentation (SNUSE) @ USC AUV Lab @ MIT Sea Grant AUVs http://auvlab.mit.edu/ Ocean Engineering @ FAU Advanced marine systems http://www.oe.fau.edu/research/ams.html AOSN Autonomous ocean samplng networks http://www.mbari.org/aosn/ Acoustic Research Laboratory (ARL) Underwater Acoustic Communications http://www.arl.nus.edu.sg/web/research/ acomms ACME Acoustic communication network for http://flipper.ncl.ac.uk/acme/ monitoring underwater environments in costal areas Underwater Acoustic Research Group @ Underwater communications http://sonar-fs.lboro.ac.uk/ Loughborough University Underwater Technologies Laboratory @ Integrated sustained ocean observing http://my.fit.edu/swood/subsea.html Florida Tech system Underwater Research LabÕs @ Simon Fraser Underwater acoustics, sonar, bottom http://www.ensc.sfu.ca/research/url/ University imaging, bottom and water column surveys with AUVs, signal processing and target detection Autonomous Undersea Systems Institute Applications of AUVs, platforms and http://www.ausi.org/ (AUSI) sensors 270 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 performance than incoherent systems for long- • Research is needed on design of low-complexity haul transmissions on horizontal channels until sub-optimal filters characterized by rapid con- ISI compensation via decision-feedback equalizers vergence, to enable real-time underwater com- for optimal channel estimation was implemented munications with decreased energy expenditure. [47]. However, these filtering algorithms are com- • There is a need to overcome stability problem in plex and not suitable for real-time communica- the coupling between the phase locked loop tions, as they do not meet real-time constraints. (PLL) and the decision feedback equalizer Hence, sub-optimal filters have to be considered, (DCE). but the imperfect knowledge of the channel im- pulse response that they provide leads to channel estimation errors, and ultimately to decreased performance. 6. Data link layer Another promising solution for underwater communications is the orthogonal frequency divi- In this section we discuss techniques for multi- sion multiplexing (OFDM) spread spectrum tech- ple access in UW-ASNs and present open research nique, which is particularly efficient when noise is issues to address the requirements of the data link spread over a large portion of the available band- layer in an underwater environment. Channel ac- width. OFDM is frequently referred to as multi- cess control in UW-ASNs poses additional chal- carrier modulation because it transmits signals lenges due to the peculiarities of the underwater over multiple sub-carriers simultaneously. In par- channel, in particular limited bandwidth, and high ticular, sub-carriers that experience higher SNR, and variable delay. are allotted with a higher number of bits, whereas Frequency division multiple access (FDMA) is less bits are allotted to sub-carriers experiencing not suitable for UW-ASNs due to the narrow attenuation, according to the concept of bit load- bandwidth in UW-A channels and the vulnerabil- ing, which requires channel estimation. Since the ity of limited band systems to fading and multi- symbol duration for each individual carrier in- path. creases, OFDM systems perform robustly in severe Time division multiple access (TDMA) shows a multi-path environments, and achieve a high spec- limited bandwidth efficiency because of the long tral efficiency. time guards required in the UW-A channel. In Many of the techniques discussed above require fact, long time guards must be designed to account underwater channel estimation, which can be for the large propagation delay and delay variance achieved by means of probe packets [30]. An accu- of the underwater channel, discussed in Section 4, rate estimate of the channel can be obtained with a in order to minimize packet collisions from adja- high probing rate and/or with a large probe packet cent time slots. Moreover, the variable delay size, which however result in high overhead, and in makes it very challenging to realize a precise syn- the consequent drain of channel capacity and chronization, with a common timing reference, energy. which is required for TDMA. Carrier sense multiple access (CSMA) prevents 5.1. Open research issues collisions with the ongoing transmission at the transmitter side. To prevent collisions at the recei- In order to enable physical layer solutions spe- ver side, however, it is necessary to add a guard cifically tailored to underwater acoustic sensor net- time between transmissions dimensioned accord- works, the following open research issues need to ing to the maximum propagation delay in the net- be addressed: work. This makes the protocol dramatically inefficient for UW-ASNs. • It is necessary to develop inexpensive trans- The use of contention-based techniques that mitter/receiver modems for underwater rely on handshaking mechanisms such as RTS/ communications. CTS in shared medium access (e.g., MACA [31], I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 271

IEEE 802.11) is impractical in underwater, for the techniques for underwater communications in following reasons: (i) large delays in the propaga- shallow water are compared, namely direct se- tion of RTS/CTS control packets lead to low quence spread spectrum (DSSS) and frequency throughput; (ii) due to the high propagation delay hopping spread spectrum (FHSS). Although of UW-A channels, when carrier sense is used, as FHSS is more prone to the Doppler shift effect, in 802.11, it is more likely that the channel be since transmissions take place in narrow bands, sensed idle while a transmission is ongoing, since this scheme is more robust to multiple access inter- the signal may not have reached the receiver yet; ference (MAI) than DSSS. Furthermore, although (iii) the high variability of delay in handshaking FHSS is shown to lead to a higher bit error rate packets makes it impractical to predict the start than DHSS, it results in simple receivers and pro- and finish time of the transmissions of other sta- vides robustness to the near–far problem, thus tions. Thus, collisions are highly likely to occur. potentially simplifying the power control function- Many novel access schemes have been designed ality. One of the most attractive access techniques for terrestrial sensor networks, whose objective, in the recent underwater literature combines similarly to underwater sensor networks, is to pre- multi carrier transmission with the DSSS CDMA vent collisions in the access channel, thus maximiz- [30], as it may offer higher spectral efficiency ing the network efficiency. These similarities would than its single carrier counterpart, and increase suggest to tune and apply those efficient schemes in the flexibility to support integrated high data rate the underwater environment; on the other hand, applications with different quality of service the main focus in medium access control in terres- requirements. The main idea is to spread each data trial wireless sensor networks is on energy-latency symbol in the frequency domain by transmitting tradeoffs. Some proposed schemes aim at decreas- all the chips of a spread symbol at the same time ing the energy consumption by using sleep sched- into a large number of narrow subchannels. This ules with virtual clustering. However, these way, high data rate can be supported by increasing techniques may not be suitable for an environment the duration of each symbol, which drastically where dense sensor deployment cannot be as- reduces ISI. sumed, as discussed in Section 2. Moreover, the In conclusion, although the high delay spread additional challenges due to the underwater chan- which characterizes the horizontal link in under- nel, such as variable and high propagation delays, water channels makes it difficult to maintain syn- and very limited available bandwidth, further chronization among the stations, especially when complicate the medium access problem in under- orthogonal code techniques are used [30], CDMA water environments. is a promising multiple access technique for under- Code division multiple access (CDMA) is quite water acoustic networks. This is particularly true robust to frequency selective fading caused by in shallow water, where multi-paths and Dopp- underwater multi-paths, since it distinguishes ler-spreading play a key role in the communication simultaneous signals transmitted by multiple de- performance. vices by means of pseudo-noise codes that are used In [41], a protocol is proposed for networks for spreading the user signal over the entire avail- with autonomous underwater vehicles. The pro- able band. This allows exploiting the time diversity posed scheme is based on organizing the network in the UW-A channel by leveraging Rake filters in multiple clusters, each composed of adjacent [43] at the receiver. These filters are designed to vehicles. Inside each cluster, TDMA is used with match the pulse spreading, the pulse shape and long band guards, to overcome the effect of prop- the channel impulse response, so as to compensate agation delay in underwater. In this case, TDMA for the effect of multi-path. CDMA allows reduc- is not highly inefficient since vehicles in the same ing the number of packet retransmissions, which cluster are close to one another. Hence, the effect results in decreased battery consumption and in- of propagation delay is limited. Interference creased network throughput. For example, in among different clusters is avoided by assigning [22], two code-division spread-spectrum access different spreading codes to different clusters. The 272 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 proposed protocol sketches also some mechanisms • It is necessary to design low-complexity encod- to reorganize clusters after node mobility. ers and decoders to limit the processing power In order to meet a required bit error rate at the required for forward error correction (FEC) data link layer of the deployed underwater sensor functionalities. Researchers should evaluate networks, it is mandatory to provide error control the feasibility and the energy-efficiency of non- functionalities for the transmitted data, since path convolutional error control coding schemes. loss and multi-path fading affecting UW-A chan- • Distributed protocols should be devised to nels lead to high bit error rates (on the order of reduce the activity of a device when its battery 10À2–10À5 [48,44]). While automatic repeat request is depleting without compromising on network (ARQ) techniques appear not to be suitable for availability. the underwater environment, because they incur a high latency, additional energy cost, and signaling overhead due to retransmissions; forward error 7. Network layer correction (FEC) techniques can be effectively employed in such an environment. The objective The network layer is in charge of determining of these techniques is to protect data by introduc- the path between a source (the sensor that samples ing redundant bits in the transmission so that the a physical phenomenon) and a destination node receiver can correct detected bit errors. This (usually the surface station). In general, while way retransmissions are not necessary although many impairments of the underwater acoustic both the transmitter and the receiver incur addi- channel are adequately addressed at the physical tional processing power drain for encoding and and data link layers, some other characteristics, decoding, respectively. There is a trade-off between such as the extremely long propagation delays, the robustness of the adopted FEC technique, are better addressed at the network layer. which depends on the amount of redundant bits in- In the last few years there has been an intensive jected in the channel, and the channel efficiency. A study in routing protocols for ad hoc wireless net- possible solution to maximize the underwater chan- works [7] and sensor networks [9]. However, due nel efficiency such a way to effectively exploit its to the different nature of the underwater environ- scarce bandwidth would be to dynamically choose ment and applications, there are several draw- the optimal amount of redundant bits according backs with respect to the suitability of the to measurements of the state of the underwater existing solutions for underwater acoustic net- channel. works. The existing routing protocols are usually divided into three categories, namely proactive, 6.1. Open research issues reactive and geographical routing protocols:

In order to enable data link layer solutions spe- • Proactive protocols (e.g., DSDV [36], OLSR cifically tailored to underwater acoustic sensor net- [26]). These protocols attempt to minimize the works, the following open research issues need to message latency induced by route discovery, be addressed: by maintaining up-to-date routing information at all times from each node to every other node. • In case CDMA is adopted, which we strongly This is obtained by broadcasting control pack- advocate, it is necessary to design access codes ets that contain routing table information with high auto-correlation and low cross-corre- (e.g., distance vectors). These protocols pro- lation properties to achieve minimum interfer- voke a large signaling overhead to establish ence among users. This needs to be achieved routes for the first time and each time the net- even when the transmitting and receiving nodes work topology is modified because of mobility are not synchronized. or node failures, since updated topology infor- • Research on optimal data packet length is mation has to be propagated to all the nodes needed to maximize the network efficiency. in the network. This way, each node is able to I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 273

establish a path to any other node in the net- Some recent papers propose network layer pro- work, which may not be needed in UW-ASNs. tocols specifically tailored to underwater acoustic For this reason, proactive protocols are not networks. In [50], a routing protocol is proposed suitable for underwater networks. that autonomously establishes the underwater net- • Reactive protocols (e.g., AODV [35], DSR [28]). work topology, controls network resources and A node initiates a route discovery process only establishes network flows. The protocol relies on when a route to a destination is required. Once a centralized network manager running on the sur- a route has been established, it is maintained by face station. The manager implements network a route maintenance procedure until it is no management and routing agents that periodically longer desired. These protocols are more suit- probe the nodes to estimate the channel character- able for dynamic environments but incur a istics. This information is exploited by the man- higher latency and still require source-initiated ager to establish efficient data delivery paths in a flooding of control packets to establish paths. centralized fashion, which allows avoiding conges- Thus, both proactive and reactive protocols tion and providing forms of quality of service incur excessive signaling overhead due to their guarantee. The performance evaluation of the pro- extensive reliance on flooding. Reactive proto- posed mechanisms has not been thoroughly car- cols are deemed to be unsuitable for UW-ASNs ried out yet. as they also cause a high latency in the estab- In [17], a framework is provided for 3D position lishment of paths, which may be even amplified based routing in ad hoc networks. It is assumed underwater by the slow propagation of acoustic that each node knows its 3D position and the po- signals. Furthermore, links are likely to be sition of the destination node, and a cell structure asymmetrical, due to bottom characteristics is leveraged in order to aggregate the topological and variability in sound speed channel. Hence, information at each node. Although it is claimed protocols that rely on symmetrical links, such that the mechanism can be applied to ocean sensor as most of the reactive protocols, are unsuited networks, all the experiments performed assume for the underwater environment. Moreover, radio frequency communications among terrestrial the topology of UW-ASNs is unlikely to vary mobile devices. dynamically on a short time scale. In [44], it is shown with simple acoustic propa- • Geographical routing protocols (e.g., GFG [12], gation models [13] that multi-hop routing saves en- PTKF [33]). These protocols establish source– ergy in underwater networks with respect to single destination paths by leveraging localization hop communications, especially with distances in information, i.e., each node selects its next the order of some kilometers. Based on this, a sim- hop based on the position of its neighbors and ple ad hoc underwater network is designed and of the destination node. Although these tech- simulated, where routes are established by a cen- niques are very promising, it is still not clear tral manager based on neighborhood information how accurate localization information can be gathered by all nodes by means of poll packets. obtained in the underwater environment with In general, while most developed protocols for limited energy expenditure. In fact, fine-grained terrestrial ad hoc networks, mostly due to scalabil- localization usually requires strict synchroniza- ity and mobility concerns, are based on packet tion among nodes, which is difficult to achieve switching, i.e., the routing function is performed underwater due to the variable propagation separately for each single packet and paths are delay. In addition, global positioning system dynamically established, virtual circuit routing (GPS) receivers, which may be used in terres- techniques can be considered in UW-ASNs. In trial systems to accurately estimate the geo- these techniques, paths are established a priori be- graphical location of sensor nodes, do not tween each source and sink, and each packet fol- work properly underwater. In fact, GPS uses lows the same path. This may require some form waves in the 1.5 GHz band and those waves of centralized coordination, and implies a less flex- do not propagate in water. ible architecture, but allows exploiting powerful 274 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 optimization tools on a centralized manager (e.g., network layer. Moreover, credible simulation the surface station) to achieve optimal perfor- models and tools need to be developed. mance at the network layer (e.g., minimum delay • Algorithms and protocols need to be developed paths, energy efficient paths), with minimum com- that detect and deal with disconnections due to munication signaling overhead. failures, unforeseen mobility of nodes or battery Furthermore, routing schemes that account for depletion. These solutions should be local so as the 3D underwater environment need to be de- to avoid communication with the surface sta- vised. Especially, in the 3D case the effect of cur- tion and global reconfiguration of the network, rents should be taken into account, since the and should minimize the signaling overhead. intensity and the direction of currents are depen- • Local route optimization algorithms are needed dent on the depth of the sensor node. Thus, under- to react to consistent variations in the metrics water currents can modify the relative position of describing the energy efficiency of the underwa- sensor devices and also cause connectivity holes, ter channel. These variations can be caused by especially when ocean-column monitoring is per- increased bit error rates due to acoustic noise, formed in deep waters. or relative displacement of communicating nodes due to variable currents. 7.1. Open research issues • Mechanisms are needed to integrate AUVs in underwater networks and to enable commu- There exist many open research issues for the nication between sensors and AUVs. In par- development of efficient routing solutions for ticular, all the information available to underwater acoustic sensor networks, as outlined sophisticated AUV devices (trajectory, localiza- below: tion) could be exploited to minimize the signal- ing needed for reconfigurations. • There is a need to develop algorithms to pro- • In case of geographical routing protocols, it is vide strict or loose latency bounds for time crit- necessary to devise efficient underwater location ical applications. To this respect, it should be discovery techniques. considered that while the delay for an acoustic signal to propagate from one node to another mainly depends on the distance of the two 8. Transport layer nodes, the delay variance also depends on the nature of the link, i.e., the delay variance in hor- The transport layer of UW-ASNs is a totally izontal acoustic links is generally larger than in unexplored area. In this section we discuss the fun- vertical links due to multi-paths [45]. damental challenges for the development of an effi- • For delay-tolerant applications, there is a need cient reliable transport layer protocol which to develop mechanisms to handle loss of con- addresses the requirements of UW-ASNs. We also nectivity without provoking immediate retrans- discuss some existing reliable data transport solu- missions. Strict integration with transport and tions for wireless sensor networks, along with their data link layer mechanisms may be advanta- shortcomings in the underwater environment. geous to this end. Noticeably, in sensor networks, reliable event • It is necessary to devise routing algorithms that detection at the sink should be based on collective are robust with respect to the intermittent con- information provided by source nodes and not on nectivity of acoustic channels. The quality of any individual report from each single source [8]. acoustic links is highly unpredictable, since it Hence, conventional end-to-end reliability defini- mainly depends on fading and multi-path, tions and solutions can be inapplicable in the which are hard phenomena to model. underwater sensor field, and could lead to waste • Accurate modeling is needed to better under- of scarce sensor resources. On the other hand, stand the dynamics of data transmission at the the absence of a reliable transport mechanism alto- I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 275 gether can seriously impair event detection due to Furthermore, due to the unreliability of the the underwater challenges. Thus, the UW-ASN acoustic channel, it is necessary to distinguish be- paradigm necessitates a new event transport reli- tween packet losses due to the high bit error rate ability notion rather than the traditional end-to- of the acoustic channel, from those caused by end approaches. packets being dropped from the queues of sensor A transport layer protocol is needed in UW- nodes due to network congestion. Most TCP ASNs not only to achieve reliable collective trans- implementations, which are designed for wired port of event features, but also to perform flow networks, assume that congestion is the only cause control and congestion control. The primary objec- for packet loss. Due to this assumption, when a tive is to save scarce sensor resources and increase packet loss occurs, they reduce the transmission the network efficiency. A reliable transport proto- rate to avoid injecting more packets in the net- col should guarantee that the applications be able work. Conversely, in UW-ASNs as in terrestrial to correctly identify event features estimated by wireless networks, it is important to discriminate the sensor network. Congestion control is needed losses due to impairments of the channel from to prevent the network from being congested by those caused by congestion. When congestion is excessive data with respect to the network capac- the cause of the packet loss, the transmission rate ity, while flow control is needed to avoid that should be decreased to avoid overwhelming the network devices with limited memory are over- network, while in case of losses due to bad channel whelmed by data transmissions. quality, the transmission rate should not be de- Most existing TCP implementations are creased to preserve throughput efficiency. unsuited for the underwater environment, since For these reasons, it may be necessary to devise the flow control functionality is based on a win- completely new strategies to achieve underwater dow-based mechanism that relies on an accurate flow control and reliability. esteem of the round trip time (RTT), which is Several solutions have been proposed to address twice the end-to-end delay from source to destina- the transport layer problems in terrestrial wireless tion. The underwater RTT can be modeled as a sensor networks. For example, in [8], event-to-sink stochastic variable with a high mean value, which reliable transport (ESRT) protocol is proposed to reflects the sum of the high delays on the links achieve reliable event detection with minimum en- composing the end-to-end path, and a high delay ergy expenditure. However, the ESRT mechanism variance, which reflects the sum of the high delay relies on spatial correlation among event flows variances on the composing link. This high- which may not be easily leveraged in underwater mean/high-variance RTT would affect the acoustic sensor networks. In fact, in terrestrial sen- throughput of most TCP implementations. Fur- sor networks nodes are densely deployed, and thus thermore, the high variability of the RTT would the physical readings of spatially close nodes may make it hard to effectively set the timeout of the be correlated (spatial correlation). Conversely, window-based mechanism that most current TCP underwater sensor nodes may be more expensive implementations adopt. and complex devices, and are usually more spar- Rate-based transport protocols seem also sely deployed. Hence, correlation among sensor unsuited for this challenging environment. In fact, readings from different sensors may not be signifi- although they do not adopt a window-based mech- cant in UW-ASNs. anism, they still rely on feedback control messages Transport layer functionalities can be tightly sent back by the destination to dynamically adapt integrated with data link layer functionalities in a the transmission rate, i.e., to decrease the transmis- cross-layer module. The purpose of such an inte- sion rate when packet loss is experienced or to in- grated module is to make the information about crease it otherwise. The high delay and delay the condition of the variable underwater channel variance can thus cause instability in the feedback available also at the transport layer. In fact, usu- control. ally the state of the channel is known only at the 276 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 physical and channel access sub-layers, while the 9. Application layer design principle of layer separation makes this information transparent to the higher layers. This Although many application areas for underwa- integration allows maximizing the efficiency of the ter sensor networks can be outlined, to the best of transport functionalities, and the behavior of data our knowledge the definition of an application link and transport layer protocols can be dynami- layer protocol for UW-ASNs remains largely cally adapted to the variability of the underwater unexplored. environment. The purpose of an application layer is multi- fold: (i) to provide a network management proto- 8.1. Open research issues col that makes hardware and software details of the lower layers transparent to management appli- In order to develop a new efficient cross-layer cations; (ii) to provide a language for querying the reliable protocol specifically tailored to underwa- sensor network as a whole; (iii) to assign tasks and ter acoustic sensor networks, the following issues to advertise events and data. must be studied: No efforts in these areas have been made to date that address the specific needs of the underwater • New flow control strategies need to be devised acoustic environment. A deeper understanding of in order to tackle the high delay and delay var- the application areas and of the communication iance of the control messages sent back by the problems in underwater sensor networks is crucial receivers. to outline some design principles on how to extend • New effective mechanisms tailored to the under- or reshape existing application layer protocols [10] water acoustic channel need to be developed, in for terrestrial sensor networks. order to efficiently infer the cause of packet losses. Some of the latest developments in middleware • New event transport reliability metric defini- may be studied and adapted to realize a versatile tions need to be proposed, based on the event application layer for underwater sensor networks. model and on the underwater acoustic channel For example, the San Diego Supercomputing Cen- model. ter Storage Resource Broker (SRB) [6,11] is a cli- • Optimal update policies for the sensor reporting ent-server middleware that provides a uniform rate are needed, to prevent congestion and max- interface for connecting to heterogeneous data re- imize the network throughput efficiency as well sources over a network, and accessing replicated as the transport reliability in bandwidth limited data sets. SRB provides a way to access data sets underwater networks. and resources based on their attributes and/or log- • The effects of multiple event occurrences on the ical names rather than their names or physical reliability and network performance require- locations. ments must be studied, as well as efficient mech- anisms to deal with it. • It is necessary to statistically model loss of con- 10. Implementations of underwater sensor networks nectivity events in order to devise mechanisms, to enable delay-tolerant applications tailored A few experimental implementations of under- to the specific underwater requirements. water acoustic sensor networks have been reported • Different functionalities at the data link and in the last few years. In this section we describe transport layer such as channel access, reliabil- two of them, one mainly concerned with military ity and flow control, should be jointly designed applications and the other with oceanographic and studied. A cross-layer approach is highly observations. recommended to accordingly optimize these The Front-Resolving Observational Network mechanisms and make them adaptable to the var- with Telemetry (FRONT) project at the University iability of the characteristics of the underwater of Connecticut relies on acoustic telemetry and channel. ranging advances pursued by the US Navy I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 277 referred to as ‘‘telesonar’’ technology [16]. The cation and networking for enhanced ocean Seaweb network for FRONT Oceanographic Sen- monitoring and exploration applications. We sors involves telesonar modems deployed in con- strongly advocated the use of a cross-layer ap- junction with three types of nodes, namely proach to jointly optimize the main networking sensors, gateways and repeaters. Sensors are ocean- functionalities in order to design communication ographic instruments serially connected to an suites that are adaptable to the variability of the acoustic modem. Gateways are surface buoys that characteristics of the underwater channel and opti- relay data from the subsurface network to the mally exploit the extremely scarce resources. shore. Repeaters are acoustic modems that relay data packets. In the various Seaweb/FRONT experiments, 20 sensors and repeaters have been Acknowledgement deployed in shallow water (20–60 m deep). By means of long range ocean bottom active sensors, The authors wish to thank Dr. Ozgur B. Akan, acoustic correlation current profilers (ACCP), Dr. Eylem Ekici, Vehbi C. Gungor and Mehmet sampling of the 3D water column is achieved with C. Vuran for their valuable comments that im- a 2D network architecture (Section 2). The net- proved the quality of this paper. This work was work enables sensor-to-shore data delivery and supported by the Office of Naval Research under shore-to-sensor remote control. contract N00014-02-1-0564. Researchers from different fields gathered at the Monterey Bay Aquarium Research Institute (MBARI) in August 2003 for a month-long exper- iment to quantify gains in predictive skills for prin- References cipal circulation trajectories, i.e., to study upwelling of cold, nutrient-rich water in the Mon- [1] UnderWater Sensor Networks at BWN Laboratory, Geor- terey Bay. Autonomous vehicles (AUVs, gliders), gia Institute of Technology, Available from . as well as other ships, vessels and platforms, en- [2] AUV Laboratory at MIT Sea Grant, Available from abled unexampled observational capabilities that . are reported on the experiment web site [4]. Exten- [3] Ocean Engineering at Florida Atlantic University, Avail- sive data are reported that show the variation of able from . the characteristics of the circulation of water dur- [4] Second field test for the AOSN program, Monterey Bay August 2003, Available from . [5] The Delay Tolerant Networking Research Group, Avail- able from . 11. Conclusions [6] SDSC Storage Resource Broker, Available from . [7] M. Abolhasan, T. Wysocki, E. Dutkiewicz, A review of In this paper, we presented an overview of the routing protocols for mobile ad hoc networks, Ad Hoc state of the art in underwater acoustic sensor net- Networks 2 (1) (2004) 1–22. work. We described the challenges posed by the [8] O.B. Akan, I.F. Akyildiz, Event-to-sink reliable transport peculiarities of the underwater channel with partic- in wireless sensor networks, IEEE/ACM Transactions on ular reference to monitoring applications for the Networking, in press. [9] K. Akkaya, M. Younis, A survey on routing protocols for ocean environment. We discussed characteristics wireless sensor networks, Ad Hoc Networks 3 (3) (2005) of the underwater channel and outlined future re- 325–349, in this issue. search directions for the development of efficient [10] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci, and reliable underwater acoustic sensor networks. Wireless sensor networks: A survey, Computer Networks The ultimate objective of this paper is to encour- 38 (4) (2002) 393–422. [11] C. Baru, R. Moore, A. Rajasekar, M. Wan, The SDSC age research efforts to lay down fundamental basis storage resource broker, in: Proceedings of the 1998 for the development of new advanced communica- Conference of the Centre for Advanced Studies on tion techniques for efficient underwater communi- Collaborative Research, 1998. 278 I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279

[12] P. Bose, P. Morin, I. Stojmenovic, J. Urrutia, Routing networks, in: C.E. Perkins (Ed.), Ad Hoc Networking, with guaranteed delivery in ad hoc wireless networks, Addison-Wesley, Reading, MA, 2001, pp. 139–172. ACM Wireless Networks 7 (6) (2001) 609–616. [29] R. Jurdak, C.V. Lopes, P. Baldi, Battery lifetime esti- [13] L. Brekhovskikh, Y. Lysanov, Fundamentals of Oceans mation and optimization for underwater sensor networks, Acoustics, Springer, New York, 2001. in: Sensor Network Operations, IEEE Press, New York, [14] J. Catipovic, Performance limitations in underwater acous- 2004. tic telemetry, IEEE Journal of Oceanic Engineering 15 (3) [30] D.N. Kalofonos, M. Stojanovic, J.G. Proakis, Perfor- (1990) 205–216. mance of adaptive MC-CDMA detectors in rapidly fading [15] E. Cayirci, H. Tezcan, Y. Dogan, V. Coskun, Wireless rayleigh channels, IEEE Transactions on Wireless Com- sensor networks for underwater surveillance systems, Ad munications 2 (2) (2003) 229–239. Hoc Networks, in press; doi:10.1016/j.adhoc.2004.10.008. [31] P. Karn, MACA—a new channel access method for packet [16] D.L. Codiga, J.A. Rice, P.A. Baxley, Networked acoustic radio, in: Proceedings of the 9th ARRL Computer modems for real-time data delivery from distributed Networking Conference, London, Ontario, Canada, 1990. subsurface instruments in the coastal ocean: Initial system [32] D.B. Kilfoyle, A.B. Baggeroer, The state of the art in development and performance, Journal of Atmospheric underwater acoustic telemetry, IEEE Journal of Oceanic and Oceanic Technology, in press. Engineering 25 (1) (2000) 4–27. [17] S. Das(Bit), R. Ragupathy, Routing in manet and sensor [33] T. Melodia, D. Pompili, I.F. Akyildiz, Optimal local networks—a 3d position based approach, submitted for topology knowledge for energy efficient geographical publication. routing in sensor networks, in: Proceedings of IEEE [18] R. Davis, C. Eriksen, C. Jones, Autonomous buoyancy- INFOCOMÕ04, Hong Kong SAR, PRC, March 2004. driven underwater gliders, in: G. Griffiths (Ed.), The [34] P. Ogren, E. Fiorelli, N.E. Leonard, Cooperative control Technology and Applications of Autonomous Underwater of mobile sensor networks: Adaptive gradient climbing in a Vehicles, Taylor and Francis, London, 2002. distributed environment, IEEE Transactions on Automatic [19] K. Fall. A delay-tolerant network architecture for chal- Control 49 (8) (2004) 1292–1302. lenged internets, in: Proceedings of ACM SIGCOMMÕ03, [35] C.E. Perkins, E.M. Belding-Royer, S. Das, Ad hoc on August 2003. demand distance vector (AODV) routing, IETF RFC [20] E. Fiorelli, N. Leonard, P. Bhatta, D. Paley, R. Bachma- 3561. yer, D. Fratantoni, Multi-AUV control and adaptive [36] C.E. Perkins, T.J. Watson, Highly dynamic destination sampling in Monterey Bay, in: Proceedings of IEEE sequenced distance vector routing (DSDV) for mobile Autonomous Underwater Vehicles 2004: Workshop on computers, in: Proceedings of ACM SIGCOMMÕ94, Lon- Multiple AUV Operations (AUV04), June 2004. don, UK, 1994. [21] L. Freitag, M. Stojanovic, Acoustic communications for [37] J.G. Proakis, J.A. Rice, E.M. Sozer, M. Stojanovic, regional undersea observatories, in: Proceedings of Ocean- Shallow water acoustic networks, in: J.G. Proakis (Ed.), ology International, London, UK, March 2002. Encyclopedia of Telecommunications, John Wiley and [22] L. Freitag, M. Stojanovic, S. Singh, M. Johnson, Analysis Sons, 2003. of channel effects on direct-sequence and frequency-hopped [38] J.G. Proakis, E.M. Sozer, J.A. Rice, M. Stojanovic, spread-spectrum acoustic communication, IEEE Journal of Shallow water acoustic networks, IEEE Communications Oceanic Engineering 26 (4) (2001) 586–593. Magazine 39 (11) (2001) 114–119. [23] S.A.L. Glegg, R. Pirie, A. LaVigne, A study of ambient [39] A. Quazi, W. Konrad, Underwater acoustic communica- noise in shallow water, Florida Atlantic University Tech- tion, IEEE Communications Magazine 20 (2) (1982) nical Report, 2000. 24–30. [24] M. Hinchey, Development of a small autonomous under- [40] V. Ravelomanana, Extremal properties of three-dimen- water drifter, in: Proceedings of IEEE NECECÕ04, October sional sensor networks with applications, IEEE Transac- 2004. tions on Mobile Computing 3 (3) (2004) 246–257. [25] B.M. Howe, T. McGinnis, H. Kirkham, Sensor networks [41] F. Salva-Garau, M. Stojanovic, Multi-cluster protocol for for cabled ocean observatories, Geophysical Research ad hoc mobile underwater acoustic networks, in: Proceed- Abstracts 5 (2003). ings of IEEE OCEANSÕ03, San Francisco, CA. [26] P. Jacquet, P. Muhlethaler, T. Clausen, A. Laouiti, A. [42] N.N. Soreide, C.E. Woody, S.M. Holt, Overview of ocean Qayyum, L. Viennot, Optimized link state routing protocol based buoys and drifters: Present applications and future for ad hoc networks, in: Proceedings of IEEE INMIC, needs, in: 16th International Conference on Interactive Pakistan, 2001. Information and Processing Systems (IIPS) for Meteorol- [27] J. Jalbert, D. Blidberg, M. Ageev, Some design consider- ogy, Oceanography, and Hydrology, January 2004. ations for a solar powered AUV: Energy management and [43] E.M. Sozer, J.G. Proakis, M. Stojanovic, J.A. Rice, A. its impact on operational characteristics, Unmanned Sys- Benson, M. Hatch, Direct sequence spread spectrum based tems 15 (4) (1997) 26–31. modem for underwater acoustic communication and [28] D.B. Johnson, D.A. Maltz, J. Broch, DSR: The dynamic channel measurements, in: Proceedings of OCEANSÕ99, source routing protocol for multi-hop wireless ad hoc November 1999. I.F. Akyildiz et al. / Ad Hoc Networks 3 (2005) 257–279 279

[44] E.M. Sozer, M. Stojanovic, J.G. Proakis, Underwater Outstanding Distinguished Lecturer Award for 1994. He acoustic networks, IEEE Journal of Oceanic Engineering received the 1997 IEEE Leonard G. Abraham Prize award (IEEE Communications Society) for his paper entitled ‘‘Mul- 25 (1) (2000) 72–83. timedia Group Synchronization Protocols for Integrated Ser- [45] M. Stojanovic, Acoustic (underwater) communications, in: vices Architectures’’ published in the IEEE Journal of Selected J.G. Proakis (Ed.), Encyclopedia of Telecommunications, Areas in Communications (JSAC) in January 1996. He received Wiley, New York, 2003. the 2002 IEEE Harry M. Goode Memorial award (IEEE Computer Society) with the citation ‘‘for significant and pio- [46] M. Stojanovic, J. Catipovic, J.G. Proakis, Phase coherent neering contributions to advanced architectures and protocols digital communications for underwater acoustic channels, for wireless and satellite networking’’. He received the 2003 IEEE Journal of Oceanic Engineering 19 (1) (1994) 100– IEEE Best Tutorial Award (IEEE Communicaton Society) for 111. his paper entitled ‘‘A Survey on Sensor Networks’’, published in IEEE Communication Magazine, in August 2002. He [47] M. Stojanovic, J.G. Proakis, J. Catipovic, Analysis of the received the 2003 ACM SIGMOBILE award for his significant impact of channel estimation errors on the performance of contributions to mobile computing and wireless networking. a decision feedback equalizer in multipath fading channels, His current research interests are in Sensor Networks, Inter- IEEE Transactions on Communications 43 (2/3/4) (1995) PlaNetary Internet, Wireless Networks and Satellite Networks. 877–886. [48] M. Stojanovic, J.G. Proakis, J. Catipovic, Performance of high-rate adaptive equalization on a shallow water acoustic Dario Pompili received the ‘‘Laurea’’ degree in Telecommunications Engi- channel, Journal of the Acoustical Society of America 100 neering in 2001, magna cum laude, (4) (1996) 2213–2219. from the University of Rome ‘‘La [49] R.J. Urick, Principles of Underwater Sound, McGraw- Sapienza’’. From June 2001 he has Hill, 1983. been working at the same university on the European Union IST Brahms and [50] G. Xie, J.H. Gibson, A network layer protocol for UANs Satip6 projects. In 2004 he earned from to address propagation delay induced performance limita- the University of Rome ‘‘La Sapienza’’ tions, in: Proceedings of IEEE OCEANSÕ01, vol. 4, the Ph.D. degree in System Engineer- Honolulu, HI, November 2001, pp. 2087–2094. ing. In 2003 he worked on Sensor Networks at the Broadband and [51] X. Yang, K.G. Ong, W.R. Dreschel, K. Zeng, C.S. Wireless Networking Laboratory, Mungle, C.A. Grimes, Design of a wireless sensor network Georgia Institute of Technology, Atlanta, as a visiting for long-term, in-situ monitoring of an aqueous environ- researcher. Currently he is pursuing the Ph.D. degree in Elec- ment, Sensors 2 (2002) 455–472. trical Engineering at the Georgia Institute of Technology. His main research interests are in Wireless Sensor Networks, [52] B. Zhang, G.S. Sukhatme, A.A. Requicha, Adaptive Underwater Acoustic Sensor Networks, and Satellite Networks. sampling for marine microorganism monitoring, in: IEEE/RSJ International Conference on Intelligent Robots and Systems, 2004. Tommaso Melodia received the ‘‘Lau- rea’’ degree in Telecommunications Engineering from the University ‘‘La Ian F. Akyildiz is the Ken Byers Dis- Sapienza’’, Rome, Italy, in 2001. He tinguished Chair Professor with the then worked on a national research School of Electrical and Computer project on Mobile Networking and Engineering, Georgia Institute of Wireless Personal Area Networks at Technology and Director of Broad- the same University. He is currently band and Wireless Networking Labo- pursuing his Ph.D. and working as a ratory. He is the Editor-in-Chief of research assistant at the Broadband Computer Networks (Elsevier) and Ad and Wireless Networking Laboratory, Hoc Networks (Elsevier) Journal. He is Georgia Institute of Technology, an IEEE FELLOW (1995), an ACM Atlanta. His main research interests are FELLOW (1996). He served as a in Wireless Ad Hoc and Sensor Networks, Wireless Sensor and National Lecturer for ACM from 1989 Actor Networks, Underwater Acoustic Sensor Networks, Per- until 1998 and received the ACM sonal and Mobile Communications.