Underwater communications and networking have been extensively studied in [4,10,16–21]. The study in [16] demonstrated the chronological development of different modulation techniques for underwater communications, and discussed the achievable data rates and transmission ranges. The authors of [10] showed that the data rate of 500 kbps can be achieved between two communicating peers with a distance of 60 m in deep water by using PSK modulation technique. Several challenging issues for underwater networking were discussed in [4] including power consumption behaviors of underwater sensor nodes, and the bizarre characteristics of underwater acoustic communication channel such as limited bandwidth, high level of multi-path effect, fading, doppler spreading, high propagation delay, and high bit error rate. The study in [17] proposed transmitter-based underwater MAC (UW-MAC) protocol combining the properties of both ALOHA and CDMA. The study in [18] has shown that multi-hop underwater communication can be more energy efficient than a single-hop communication in a long distance.

To increase the data rate over the bandwidth-limited acoustic channel, either the bandwidth, or the spectral efficiency in the unit of bits per second per Hertz (b/s/Hz), or both, need to be increased [20,21]. Toward achieving this goal, the studies in [20] showed that MIMO-OFDM technique can easily attain spectral efficiency of 3.5 bits/sec/Hz in one experiment and a data rate of 125.7 kb/s over a bandwidth of 62.5 kHz in another for an acoustic channel of distance 1,500 m.

There also have been studies on alternative underwater communication technologies for short-range, high rate data transmissions [19,22–25]. For example, it was shown in [22] that 1 Gbps data rate can be achieved for the distance of two meters using underwater optical link. In [23], integrating suitable MODEM using omnidirectional optical transceiver, a preliminary design of optical communication system was carried out to realize 10 Mbps data rate between two underwater sensor nodes. Also, in [19], authors not only discussed the short-range, high-bandwidth optical communication link between an AUV and a fixed sensor station, but also proposed underwater communications using low-cost light emitting diodes (LEDs). The authors of [24] studied electric current based underwater communications, and showed that 1 Mbps data rate is achievable for the short distance (e.g., 10 m) communications.

The possible integration of a short-range optical communication technology and long range underwater acoustic communications is studied in [25]. This communication system can be used for an AUV to communicate with fixed underwater sensors with a high data rate, which can justify our assumption of short-range and high rate communications between AUVs and GWs. Those related studies show that our assumptions on short communication range for attaining a large amount of data transmissions is reasonable. Furthermore, the development of alternative underwater communication technologies has enhanced the possibility of a short-range and very high rate data transmission between AUVs and underwater devices.

In this section, we discuss various underwater monitoring systems and technologies.

The sink node broadcasts, at every s seconds, to the network PHE messages, which will be forwarded and used by U-sensors to determine the next hop to the sink. When the sink node receives beacon signals from an AUV, it initiates negotiation with the AUV to determine the channel for data transmission, the transmission rate, and the modulation scheme. Then, the sink node receives the aggregated data from the AUV using the negotiated short-range/high-rate channel. The sink node also receives data from U-sensors that have generated the data or are relaying data generated by other U-sensors. A sink node discards PHE messages from GWs. It is worth noting that AURP can be readily extended for an architecture with multiple sink nodes.

U-sensor nodes collect data using equipped sensors and periodically send the data to the sink. U-sensors determine the next hop to forward the data packet based on the information in the PHE messages that they have received from the sink or GWs. More specifically, when a U-sensor receives a PHE message from a GW or the sink, it stores the message in DET_NEXT_HOP_QUEUE. The U-sensor also maintains a timer, DNH _TIMER. When the timer expires, the U-sensor examines the travel time of the PHE message and the number of hops that the PHE message passed from GWs or the sink. Based on the information, the U-sensor determines the next hop that will forward the sensed data to a GW or the sink, such that the data can be delivered to GWs/sink with the minimum path length and delay.

Other routing metrics such as ETX (Expected Transmission Count) and ETT (Expected Transmission Time) can be also applied for path selection.

Similarly to the sink node, GW nodes periodically flood PHE messages that will be used by U-sensors. When a GW receives data packets destined to the sink from the U-sensors, it stores them in a queue until it hears a beacon signal from an AUV. Upon reception of a beacon signal from an AUV, the GW and the AUV negotiate on the communication channel and send the stored data to the AUV using the selected channel.

If a GW node receives PHE messages generated by other GWs or the sink node, it simply discards the messages.

The GWs can be pre-determined or dynamically selected by AUVs. In case the positions of GWs are pre-determined, more optimal trajectories of AUVs can be applied and dedicated and more powerful U-sensor nodes can be used as GWs.

Multiple AUVs move around the survey area to collect data from GWs and forward it to the sink when it passes by the sink. Each AUV periodically broadcasts a beacon signal to show its presence to GW nodes and the sink node. AUVs communicate with GWs and the sink node using a separate short-range, high rate data channel. As a result, AUVs, GWs, and the sink form a separate collision domain for the medium access from the remaining network (e.g., U-sensors and GWs), which significantly reduce the packet collisions in a multi-hop network.

If an AUV has failed (e.g., due to communication or mechanical malfunction) the data it carries can be lost. One possible way of recovering from this loss is that the GW keeps the data, after it sends data to an AUV, until it receives acknowledgement (ACK) from sink via the AUV at the next AUV’s visit. If the GW does not receive ACK from the AUV, it assumes that the data was lost and retransmits it to the AUV. Another point to be considered is how fast the failure can be detected to minimize the recovery delay. Fast failure detection can be achieved since the mothership usually keeps track of the AUV.

As the underwater environment is remarkably different from its terrestrial counterpart [4,35], the existing wireless modules cannot be deployed without specific re-engineering. The internodal distance as well as the link orientation of the underwater network highly affects the underwater channel characteristics [35]. Thus the underwater channel is attributed as the combination of the worst aspect of terrestrial mobile for poor quality and satellite radio channel for high latency [8]. Therefore, it is crucial to accurately emulate underwater channel to warrant the outcome of a simulation process. A lot of research in this context [41,42] have tried to include the impact of loss and delay of the underwater communication into their acoustic channel design.

In order to accurately emulate the underwater communication environment for simulation, we have used an underwater acoustic channel model designed in [43]. In [43], the absorption at a given frequency is approximated based on the Thorp’s approximation shown in [44]. To estimate the total attenuation and eventually obtain the SNR that is used by the NS2 simulator, the model in [43] also incorporates spreading loss and various underwater noises, such as turbulence, shipping, wind, and thermal noises.

The authors of [43] also modeled another important underwater communication factor, the underwater acoustic propagation speed based on [45]. Therefore, in our simulation, the propagation speed s_p in meter per seconds is calculated according to the model shown in [43,45], which is as follows s p = 1449.05 + 45.7 t − 5.21 t 2 + 0.23 t 3 + ( 1.333 − 0.126 t + 0.009 t 2 )   ( S − 35 ) + 16.3 z + 0.18 z 2where t, z, and S represents one tenth of the water temperature, the depth and the salinity of the water, respectively.

For example, we obtain approximately 0.33 s as a propagation delay over a distance of 500 m at a depth of 2,000 m from the surface level.

In order to gain an insight of the effects of AUVs’ movement on the performance of underwater network, we design simple yet practical AUV movement patterns, where AUVs move along an elliptical trajectory. More specifically an AUV trajectory is defined by an equation of ellipse, ( x − x c ) 2 a 2 + ( y − y c ) 2 b 2 = 1, where a, b, x_c, and y_c are real numbers. Another variable θ is used to denote the angle of rotation from the positive Y-axis.

Note that the network simulator NS2 only supports a linear movement between two positions at a given constant speed. In order to make the designed AUVs’ movement pattern work with the NS2 simulator, an elliptical trajectory is partitioned into a large number of small line segments that are represented by a list of (x, y) positions. Assuming an ellipse is centered at (0, 0) and θ = 0 in order to facilitate the discussion, the position of AUV at time t, (x_t, y_t), is obtained by using following equation. (1) x t 2 a 2 + y t 2 b 2 − 1 − ( x t − x t − 1 ) 2 + ( y t − y t − 1 ) 2 + Δ t * v = 0where (x_t−1, y_t−1), Δt, and v are the previous position of the AUV at time t − 1, the unit time in the NS2 mobility trace, and the average speed of AUV respectively. Note that between two solutions of Eq. (1), only one is selected as the next position that corresponds to the current move direction of the AUV.

Figures 3 and 4 show the AUVs’ movement pattern in an area of 25,000 m × 25,000 m used for the performance study. In Figure 3, one AUV is considered with four different parameter sets for elliptical trajectories, Traj(i) where 1 ≤ i ≤ 4. The sink node is assumed to be located at the bottom point of the each path. In Figure 4, multiple AUVs’ trajectories are shown. Let each movement pattern be denoted by (a, b, θ). Then, Figure 4(a) shows the trajectory with one AUV, where (a, b, θ) = (600, 800, 0). Two AUV case is shown Figure 4(b), where (a, b, θ) = {(500, 800, π 6), (500, 800, − π 6)}. Figure 4(c) show trajectories of three AUVs, where (a, b, θ) = {(400, 800, π 4), (400, 800, 0), (400, 800, − π 4)}. Finally, four AUVs’ trajectories are shown in Figure 4(d), where (a, b, θ) = {(300, 800, 3 π 10), (300, 800, − π 10), (300, 800, − π 10), (300, 800, − 3 π 10)}.

It is worthwhile to note that the trajectories of AUVs can be better optimized with the topology information, in particular, the positions of GWs. More specifically, both positions of the GWs and the trajectories of AUVs can be calculated before deployment such that the performance of the underwater sensors network is maximized.

In order to test the routing scheme proposed in Section 4, we use NS2 simulator and also incorporate the underwater channel model developed by the authors of [43]. U-sensors and AUVs are deployed within a 2,500 m × 2,500 m underwater region. The range of numbers of U-sensors and AUV are (20,80) and (1,4), respectively. The default value of the number of U-sensors is set to 50. The depth of water is assumed to be 2,000 m, which leads to around 0.16 s propagation delay over a distance of 250 m. To obtain more practical simulation results, the specification of a commercial acoustic modem, UWM2200 manufactured by LinkQuest Inc. [46], is used. For example, the transmit and receive power of the acoustic interface are set to 6 W and 1 W respectively, and the maximum data rate of the acoustic interface is set to 14 Kbps. It is assumed that each U-sensor periodically (every 30 s) sends data of 100 Bytes to the sink. GWs and the sink broadcast PHE messages at every 150 and 60 s, respectively. AUVs broadcast a beacon signal at every 20 s, and it is assumed that AUVs can be periodically recharged or replaced during the underwater monitoring operation. Thus, in this work, we assume that only energy consumption of U-sensors affect the network lifetime. The average speed of AUVs set to 9.7 knots, which can be achieved by latest AUVs (e.g., AUV developed in [47] can move with a speed of 15 knots).

In order to verify AURP, we compare it with a sensor network routing protocol, referred as Grad hereafter, which is based on interests and gradients used in Directed Diffusion [11]. In Grad, The sink periodically floods Interests to the network, which results in formation of Gradient in the opposite direction to the flow of the Interests. The source nodes and relaying nodes use the Gradient to deliver data.

Figure 5 shows the effect of the number of U-sensors on the network performance. As many as 50 U-sensors are randomly deployed in the area, and one AUV is used. AURP with different numbers of GWs, from 1 to 3, is compared with Grad. As shown in Figure 5(a), AURP and Grad show a low delivery ratio when the number of U-sensors is 20. The reason of a low delivery ratio with 20 U-sensors is that there are not enough U-sensors to form a connected network. In a partitioned network, as shown in Figure 5(a), Grad shows a very poor performance since a large portion of the sensing data can not be relayed to the sink. Meanwhile, AURP can achieve a relatively higher delivery ratio even in a partitioned network. When there are 40 U-sensors, AURP achieves over 0.9 delivery ratio, while Grad achieves only about 0.5 delivery ratio. As the number of U-sensors grows, the delivery ratios of AURP and Grad decrease, since the network becomes congested due to the high network traffic. Note that the delivery ratio of AURP is much higher than Grad even in a dense network. For instance, with 80 U-sensors, AURP shows the delivery ratio in the range of (0.78, 0.88) depending on the number of GW, which is higher than twice of Grad case. The results shown in Figure 5(a) also implies that deploying more GWs can improve the delivery ratio.

In a network using AURP, the U-sensors consume less energy over different number of U-sensors than U-sensors in a network using Grad, as shown in Figure 5(b). The gap of energy consumption becomes greater as the number of U-sensors increases. Note that, in this work, the energy consumption of U-sensors is of interest, hence only U-sensors’ consumed energy is collected.

As shown in Figure 5(c), the packet delay in AURP is much higher than that of Grad, since a large portion of packets are carried by AUVs to be delivered to the sink. In other words, the packet delay mainly depends on the travel distance and the speed of AUV. Even though the delay can be reduced by optimizing the trajectory and increasing the speed of AUVs, the store-carry-forward process will lead to an inevitable packet delay. Therefore, the proposed network architecture will be suitable for an underwater monitoring system that requires high delivery ratio, low energy consumption, and allows the packet delay of several hundred seconds. Although this delay may pose a limitation in applying the network to some military/tactical applications that requires strictly real time data transmissions, the proposed architecture can be used for most applications that can tolerate several minute delay yet require high rate and reliable data transmissions. The packet delay results in other scenarios are similar due to the fact that the packet delay mostly depends on the AUVs’s speed and trajectories rather than network situation. Therefore, from now on we focus on the data delivery ratio and energy consumption as performance metrics.

Figure 6 shows the effect of the network load on the performance. The × axis represents the sensing intervals from 50 s to 10 s. As shown in Figure 6(a), AURP shows a much higher data delivery ratio than Grad in all cases. The figure also shows that the performance is improved with a larger number of GWs. Moreover, the difference of delivery ratio values becomes larger when the network load is high. As shown in Figure 6(a), when the sensing interval is 10 s, the data delivery ratio of AURP with three GWs is around 0.9, while that of Grad is less than 0.2. It is because, when the network load is high, the transmissions in Grad experience network congestion which results in a high level of packet collision. On the other hand, in AURP, the increase of the number transmissions is much lower than that of Grad due to less number of hops between U-sensors and the sink. The total energy consumption of the U-sensors using AURP is lower than those with Grad. As shown in Figure 6(b), Grad based U-sensors consume more energy for data transmission and reception than U-sensors using AURP. This result also comes from the fact that U-sensors using AURP transmit less packets.

A higher performance of AURP can be achieved by using more than one AUV. Figure 7 shows the effect of multiple AUVs on the delivery ratio and energy consumption. As the number of AUV increases, a higher delivery ratio can be achieved. For example, AURP achieves 0.97 delivery ratio when (#AUVs, #GWs) = (3,3), while it shows 0.85 delivery ratio when (#AUVs, #GWs) = (1,3) as shown in Figure 7(a). Note that #GWs represents the number of GWs for each AUV; if (#AUVs, #GWs) = (3,3), then there are 9 GWs in total. Grad has a constant value, approximately 0.5, over the simulations since it does not use AUVs. The number of AUVs also has a positive effect on the energy consumption. As shown in Figure 7(b), the U-sensors use less energy when there are more AUVs, since U-sensors can send data to AUVs with a shorter path.

The trajectories that AUVs follow also affect the network performance as shown in Figure 8. Four trajectories, Traj₁,…, Traj₄, shown in Figure 3 are used. The X-axis represents the value of a shown in Figure 3. As shown in Figure 8(a), Traj₂ and Traj₃ show a better data delivery ratio. This is because the number of transmissions due to multihop relay becomes larger when an AUV follows a too small or a too large trajectory, which implies that selecting the trajectory of AUVs is also important to maximize the network performance.