Latency-Optimized and Energy-Efficient MAC Protocol for Underwater Acoustic Sensor Networks: A Cross-Layer Approach

Standard acoustic transducers cannot simultaneously transmit and receive. On space-constrained autonomous underwater vehicles (AUVs) and compact stationary nodes, transducers in different frequency bands cannot be spatially separated far enough to provide full-duplex connections. Underwater network communications are therefore almost always half-duplex. Furthermore, transducer sizes are proportional to wavelength, and due to space constraints, small AUVs are often restricted to using higher center frequencies, generally above 10 kHz [16]. Another issue is that it is easy for small AUVs to transmit at high data rates but harder for them to receive at high rates. The two main reasons for this asymmetry nature are [17] the following.

For sending short commands to underwater vehicles, low-rate and moderate-rate transmission methods, such as FSK, have been shown to be relative reliable. While, when allowing for transfer of complicated mission plans or other large data commodities, high-rate methods are needed. In [19], a two-way phase-coherent communication scheme is proposed to provide high-rate connectivity for AUVs. The telemetry system designer is driven to employ a coherent system for greater data throughput, but the sensitivity issues of coherently demodulated systems to channel fluctuations may require more robust incoherent demodulation. The selection must be made careful consideration of anticipated channel behavior and coherent signaling used only in applications where the fluctuation level and dynamics are sufficiently low to permit coherent carrier acquisition and tracking.

The propagation speed in the underwater channel is five orders of magnitude lower than that of the radio channel. Large delays between transactions can reduce the throughput the system considerably if it is not taken into account. Another crucial challenge for UASNs designers is to develop a system that will run for years unattendedly, which calls for not only robust hardware and software, but also lasting energy sources. However, currently, sensor nodes are powered by battery, whose available energy is limited. Therefore, protocols and applications designed for UASNs should be highly efficient and optimized in terms of energy.

Underwater acoustic networks are generally formed by acoustically connected ocean-bottom sensors, autonomous underwater vehicles, which are robots and can travel underwater, and surface stations. While many applications require long-term monitoring of the deployment area, the battery-powered network nodes limit the lifetime of UASNs. In this paper, we propose an underwater acoustic sensor network with mobile agent (UASNMA), a new network architecture especially for low power and large-scale underwater networks. As illustrated in Figure 1, a group of sensor nodes are anchored to the bottom of the ocean with deep ocean anchors. AUVs are exploited as mobile agent in our network. These devices are powered by batteries or fuel cells and can operate in water as deep as 6000 meters. Advances in propulsion systems and power source technology give these robotic submarines extended endurance in both time and distance. Compared to fixed sensor nodes, AUVs are powerful hardware units, both in their communication, processing capability, and in their ability to traverse the networks.

For large-scale networks, there are three basic topologies that can be exploited to interconnect network nodes: centralized, distributed, and multihop topologies [20]. Considering scalability, power consumption, and robustness, we utilize a distributed topology for our UASNMA. That is, the distributedly deployed fixed sensor nodes collect environment-related data, and then transmit those data to AUVs around them. Through the high rate connections between AUVs and surface stations or satellites, which can further be connected to a backbone, such as the Internet, through a RF link, collected data are relayed to the data centers with short transmission delay. AUVs need not always to be present or operational along with individual sensor nodes for burst traffic, and they are in action only when it is necessary to collect data and perform network maintenance. Through AUVs traversing the network, the computing complexity and energy for routing is reduced. This configuration creates an interactive environment where scientists can extract real-time data from multiple distant underwater instruments. After evaluating the obtained data, control messages can be sent to individual instruments and the network can be adapted to changing situations.

For our network model, to transmit an -symbol message for a distance , the node expends

and to receive this message, the node expends

The typical transmission electronics energy, , is 50 W, and the typical value for reception, , is 0.2 W to 2 W [16]. The electronics energy, , depends on factors such as coding, modulation, pulse-shaping, and matched filtering. The transmitter outputs to traverse a unit distance and reach the receiver with an acceptable SNR, which depends on the distance to the receiver and the acceptable bit error rate. Hence, the total energy consumption with fixed sensors is

However, for the flat ad hoc architecture, we use the same model in [21] to calculate the expected total energy expenditure as follows

where and are the expected energy consumed by the one-hope transmission and reception of a sensor, which is at a distance away from the access point and the expected number of hops that sensor's packet needs to make to reach to the access point.

Therefor, for a network with fixed sensors, which are randomly and uniformly deployed on a disk with a radius of meters, the total energy consumption is

Combining the advantages of schedule-based MAC protocols and random-access MAC protocols, we apply energy-efficient asynchronous schedule-based MAC (ASMAC) protocol [22] to charge the common channel sharing among sensor nodes. ASMAC can remove accumulative clock-drifts without any network synchronization. Moreover, ASMAC adjusts essential algorithm parameters, which can ensure adequate successful transmission rate, short waiting-time, and high energy-utilization adaptively according to the traffic strength and network density.

ASMAC protocol divides system time into four phases: T RFR-P hase, S chedule-P hase, O n-P hase, and O ff-P hase (Figure 2).

At the end of each O n-P hase, nodes go to "vacation"—O ff-P hase, for a period of time. Thus, new arrivals during an O n-P hase can be served in first-in-first-out (FIFO) order. While, new arrivals during an O ff-P hase, rather than going into service immediately, wait until the end of this O ff-P hase, then they are served in O n-P hase and in FIFO order. Interarrival time and service time for data packets are independent and follow general distribution and individually. For average interarrival time , we have . Similarly, for average service time , we have .

According to received schedule messages (Figure 4), nodes set up their own phase-switching schedules, which ensure them to switch to the same phase simultaneously.

ASMAC utilizes two techniques to make its scheme robust and feasible to use free-running timing method [23], which allows nodes to run on their own clocks and makes contribution to save the energy used by setting up and maintaining the global or common timescale. Firstly, schedule messages are broadcasted. Leveraging the property of broadcast, schedule messages can reach all data-collection nodes at the same time once ignoring the difference of propagation time of them (it is reasonable since the propagation time within a cluster, which is between 0.1 and 1 microseconds). Moreover, nodes go to O n-P hase immediately after receiving schedule messages. Secondly, in a schedule message, all time references, such as On-Duration and Off-Duration, are relative values rather than absolute values. This property can eliminate errors introduced by sending time and access time.

Note that, based on schedule messages and nodes' local clocks, phase-switching schedules are supposed to be established at each node to ensure matching operations if no clock-drift. However, mismatching operations among nodes are unavoidable, since there are always clock-drifts caused by unstable and inaccurate frequency standards. So it is possible that transmitters have powered on their radios to send message, but receivers' radios are still powered off. Those mismatching operations cause communication to fail. Moreover, with the accumulative clock-drift becoming bigger and bigger, the impact on communications turns to be more and more serious. The solution in ASMAC is to rebroadcast schedule message, which forces data-collection nodes to remove accumulative clock-drifts and to reestablish matching schedules.

While, how can data-collection nodes know the time of next schedule broadcast so as to power on their radios? The solution is that ASMAC includes reschedule interval information into schedule messages. To preestimate the value of schedule interval, a rescheduling-FLS is designed to monitor the influence of accumulative clock-drifts, the variance of traffic strength and service capability on communications. Then ASMAC can adjust schedule interval and power-on/off duration adaptively. ASMAC uses

as the interval adjustment function. Where is the interval for the th schedule broadcast, is the th adjustment factor determined by our rescheduling-FLS.

Hence each node within a cluster is synchronized to a reference packet (schedule message) that is injected into the physical channel at the same instant. Furthermore, after a same period of time specified by , all nodes switch to O ff-P hase and stay there for a period. Finally, all nodes switch back to O n-P hase. Phase is circulatedly switched like this way (see Figure 5).

Therefore, there should be at least one available AUV for each sensor cluster during its O n-P hase. According to those information, we can decide an AUV's navigation path and how many AUV are needed to cover the whole network.

Normally, shortening the transmission delay for each data packet is considered during the transmission protocol design. However, for large data applications, the transmission delay for whole set of data should be more considered, since after all information received the data processing process can be started. Therefore, choosing the transmission delay for one set of data is more reasonable/meaningful to optimize transmission delay performance for MAC protocols. Moreover, almost all underwater communications use acoustics. The speed of sound underwater is approximately 1500 m/s. This leads to unneglectable propagation delays. The transmission delay ( ) for a set of data is described as follows

where is the number of segment a set of data is divided into at MAC layer; is the length for information part of a frame. The unit is bit; is the length for MAC header of a frame. The unit is bit; is the data reception rate. The unit is bits/sec; is the distance between a sensor node and an AUVs; and is the speed of sound in water.

From (7), note that, there are two options to reduce the value for :

Assuming we know the bit error rate (BER) for the given channel and the transmission/receiption scheme (including modulation/demodualtion and channel coding/decoding, etc.), then the frame error rate (FER) can be formulated as in (8). Obviously, for the same BER channel, larger frame size will increase the FER, which means decreasing the effective throughput of systems,

Therefore, there is a tradeoff between reducing the data transmission delay and increasing the data successful transmission rate during our MAC protocol design. This also inspire us to achieve the criterion for determining frame size, as described in (9). Consequently, according to the expected FER, channel status, and transmission/receiption schemes, we can adaptively determine the optimum frame size for transmission,

Therefore, for , the longest length of payload part is when fixing the BER. Bring it into (7), the shortest transmission latency is shown in (10) when fixing . Here is the total size of a set of data

Compared to terrestrial sensor networks, the underwater channel is asymmetric. That is, the data receiving rate at the receiver side should be some lower than the one at the transmission side for propulsion noise, since the receiver's SNR is much lower than the one of transmitters. This will be a bottleneck, especially for large data applications. How to increase data reception rate, while keeping the same performance in terms of BER, is our another task in this paper.

To reduce the system-error probability, error correction coding is used. Rayleigh and Rician signal fading imposes an expensive tradeoff between system reliability and transmitted power. Convolutional codes [24] are a class of important codes due to their flexibility in code length, soft decodability (e.g., using the BCJR algorithm or soft output Viterbi algorithm (SOVA)), short decoding delay (e.g., using windowed Viterbi algorithm), and their role as component codes in parallelly/serially concatenated codes. Puncturing allows convolutional codes to flexibly change rates and is widely used in applications where high code rates are required and where rate adaptivity is desired [24].

We apply convolutional coding to encode the information bits, then the code words are interleaved. Interleaver [24] was used to eliminate the correlation of the noise/fading process affecting adjacent symbols in a received code word, but here we use interleaver to make sure that the incorrect symbols in one compromised path will be spread after deinterleaver so that the Viterbi decoder will perform well. The interleaved bits are inserted with some unique words (for demodulation purpose), and then these bits are modulated to symbols. For data receiving rate enhancement, we apply the puncture idea at the receiver side. Since the received symbols are reduced, the power per bit is increased. Therefore, the SNR at the receiver side is increased with the same environment noise and the same channel fading effects. While, puncture scheme impact the performance directly.

In general, the design of an error correction coding system usually consists of selecting a fixed code with a certain rate and all the data to be transmitted/reveived and adapted to the average or worst channel conditions to be expected. In many cases, one would like to be more flexible because the data to be transmitted/received have different error protection needs and the channel is time varying or has insufficiently known parameters. Consequently, flexible and adaptive data transmission/reception scheme is expected. Hence, at the receiver side, the puncture rate can be adaptively adjusted based on current channel state to ensure the satisfied BER for received data.