**Energy Efficient Communication for Underwater Wireless Sensors Networks**

Ammar Babiker and Nordin Zakaria *PETRONAS University of Technology Malaysia* 

## **1. Introduction**

46 Energy Efficiency in Communications and Networks

Shahab, A. (December 2010). *On Achieving the Minimum Energy for Sending One Single Bit* 

Shin, S.-K.; You, Y.-S.; Lee, S.-H.; Moon, K.-H.; Kim, J.-W.; Brooks, L. and Lee, H. S. (2008). *A* 

Steyaert, M.; Bijker, W. & Sevenhans, J. (1991). ECL-CMOS and CMOS-ECL interface in 1.2-

Sundström, T. & Alvandpour, A. Utilizing process variations for reference generation in a

Tan, C. & Chen, Z., An efficient CMOS operational amplifier for driving large capacitive

Tsividis, Y. (2002). *Mixed Analog-Digital VLSI Devices and Technology*. World Scientific, ISBN:

Van der Plas, G.; Decoutere, S. & Donnay, S. A 0.16pJ/Conversion-Step 2.5mW 1.25 GS/s 4b

van Veldhoven, R. H. M. van; Rutten, R. and Breems, L. J. (2008). An inverter based hybrid

Vaz, A.; Solar, H.; Rebollo, I.; Gutierrez, I. and Berenguer, R. (2010). *Long range, low power* 

Vittoz, E. A. (1994). Micropower techniques, In: *Design of MOS VLSI Circuits for* 

Vittoz, E. A. (2009). Weak Inversion for Ultra Low-Power and Very Low-Voltage Circuits, *IEEE Asian Solid-State Circuits Conference*, (November 2009), pp. (129-132). Wang, H.; Xu, J. and Wu, X. (2009). *A high power efficiency Class AB switched-opamp for low voltage* 

*Circuits Conference, 2006. ESSCIRC 2006*, (19-21 Sept. 2006), pp.(187-190). Yang, H.Y.; Sarpeshkar, R. (2005). *A time-based energy-efficient analog-to-digital converter*, IEEE Journal of Solid-State Circuits*,* Vol.40, No.8, (Aug. 2005), pp.(1590- 1601). Yang, H.Y.; Sarpeshkar, R. (2006). *A Bio-Inspired Ultra-Energy-Efficient Analog-to-Digital* 

Yao, L.; Steyaert, M. S. J. & Sansen, W. (2004). A 1-V 140-lA 88-dB audio sigma-delta

Yao, L.; Steyaert, M. & Sansen, W. (2006). *Low-Power Low-Voltage Sigma-Delta Modulators in Nanometer CMOS*. Springer. ISBN-13 978-1-4020-4139-6 (HB), The Netherlands. Yeknami, A.F.; Qazi, F.; Dabrowski; J.J. and Alvandpour, A. (2010). *Design of OTAs for Ultra-*

*Solid-State Circuits*, Vol. 26, No. 1, (January 1991), pp. (18–24).

*International Solid-State Circuits Conference 2006, February 2006.* 

Microwave Symposium Digest, (May 2010), pp.(836-839).

*Telecommunications.* Prentice-Hall, ISBN: 0-13-203639-8, USA.

Regular Papers*,* Vol.53, No.11, (Nov. 2006), pp.(2349-2356).

Signals and Electronics Systems ICSES, Poland, September 2010.

(November 2004), pp. (1809–1818).

http://repository.tudelft.nl/.

(364-368).

Guilin, October 2007.

9812381112, Singapore.

pp. (492–493).

*IEEE Trans. VLSI Syst.*, (Jun. 2008).

*with Feedback.* MSc Thesis from Delft University of Technology. Available from:

*fully-differential zero-crossing-based 1.2 V 10b 26 MS/s pipelined ADC in 65 nm CMOS*,

pm CMOS for 1.50-MHz digital ECL data transmission systems, *IEEE Journal of* 

Flash ADC. *IEEE transactions on circuits and systems II*, vol. 56, nº 5, May 2009, pp.

loads, International Conference on ASIC, ASICON 2007, ISBN 978-1-4244-1132-0,

ADC in a 90nm Digital CMOS Process. *Digest of Technical Papers of IEEE* 

ΣΔ modulator, IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, (Feb. 2008),

*UHF RFID analog front-end suitable for batteryless wireless sensors,* IEEE International

*low power sigma-delta modulators.* IEEE International Conference of Electron Devices and Solid-State Circuits, EDSSC 2009*.* Vol. , No., (25-27 Dec. 2009). pp.(429-432). Wismar, U.; Wisland, D. & Andreani, P. (2006). A 0.2V 0.44μW 20 kHz Analog to Digital *SD*

Modulator with 57 fJ/conversion FoM. *Proceedings of the 32nd European Solid-State* 

*Converter for Biomedical Applications*. IEEE Transactions on Circuits and Systems I:

modulator in 90-nm CMOS, *IEEE Journal of Solid-State Circuits*, Vol. 39, No. 11,

*Low-Power Sigma-Delta ADCs in Medical Applications*. International Conference on

Water covers more than 70% of the planet, contains much of its natural resources, and defines the greater territories of many nations. With the increasing use of underwater sensors for the exploitation and monitoring of vast underwater resources, underwater wireless sensor network (UWSN), mostly based on acoustic transmission technologies, have been developing steadily in terms of operation range and data throughput.

As in terrestrial sensor networks, various data transport protocols have been designed for UWSN (Pompili, 2007). However, as yet, there is no internationally accepted standard for underwater communication. The lack of standard is due to the technical challenges that still persist in establishing reliable underwater wireless data communication. Firstly, in underwater environment, electromagnetic wave is rarely of use, unlike in terrestrial space, as it can only travel a short distances due attenuation and absorptions effects. Optical signal suffers from scattering and absorption in underwater (Akyildiz, et al., 2005). Hence, to date acoustic energy is the most widely used type of signals used in underwater data transmission. Secondly, the fluctuating nature of ocean condition causes high bit error rate in acoustic transmission. Underwater acoustic transmission is also affected by path loss caused by spreading and absorption, noise which comes from many sources like water current, rain, wind, seismic and volcanic activities or biological phenomena (Pompili, 2007). Signal reflection and refraction from the surface and seabed, topographic sources like hills and hollows are some example error sources.

Hence in underwater environment, the two main issues of concern are namely: reliability and energy efficiency. These two issues are inter-twined. Reliability requires errorcorrection, and error-correction requires energy. More reliability tends to imply higher energy consumption, causing difficulty in applications that require nodes to be operated underwater for long periods of time without batteries recharging, and in aquatic environments that render hard the task of recharging or replacing batteries (Pompili, 2007; Preisig, 2007). Appropriate strategy must therefore be in-place to ensure reliable data transmission, while conserving energy.

In this chapter, we focus on the energy efficiency issue in UWSN. We develop a mathematical model of the efficiency of acoustic data communication in realistic underwater environment. We analyze existing error-correction techniques, and we then propose a new adaptive hybrid error correction technique that improves upon existing techniques.

Energy Efficient Communication for Underwater Wireless Sensors Networks 49

• Doppler spread: it is significant in underwater acoustic channel, and cause degradation

Most of the factors mentioned above are caused by the chemical-physical properties of the sea water such as temperature, salinity and density, which these factors vary with depth,

Automatic Repeat reQuest (ARQ) and Forward Error Correction (FEC) are two commonly used strategies to combat error in underwater transmission (Bin et al, 2008). ARQ which proposes retransmission (Kunal et al., 2010), are widely used in data communications system for error control as they are simple and provide high reliability. However, the throughput is not constant and decreases rapidly in high bit error rate cases (Lin et al., 1984). In FEC, redundancy is added for error prevention. Redundant bits are encapsulated with data bits to form encoded information. However this increases the payload for transmission. Addition of redundant bit is known as channel coding. Error Correcting Codes (ECC) (block or convolutional) are used for this purpose. FEC codes have constant throughput which is equal to the code rate. However it has the drawback of using parity bit irrespective of the existence of errors. Reliability can be enhanced by combining FEC and ARQ, forming what

ARQ uses error detection codes, acknowledgement and/or negative acknowledgement messages, and time out to retransmit error packet. The basic idea is that the transmitter after sending the packet waits for specific time (time out) to receive an acknowledgement. If it receives positive acknowledgement (ACK), it sends the next packet. On the other hand, if it receives negative acknowledge (NAC) or timed out before receiving any acknowledgement, it then retransmits the same packet. The process repeats until an ACK has been received by

In (Tan et al., 2007), an opportunistic (hybrid implicit/explicit) acknowledgement scheme suitable for stop and wait protocols in underwater is proposed. The simple stop and wait (S &W) protocol is chosen as it is the most popular method in underwater acoustic communication due to the half-duplex property of acoustic modem. In the context of a multi hop channel, the work in (Tan et al., 2007) proposed that the acknowledgement can be achieved explicitly by transmitting an acknowledgement packet per successfully received

In (Lee et al., 2008), the channel sharing property inherent in underwater environment is utilized in proposing an efficient ARQ scheme. In this scheme packet size is controlled in such a way that transmission time becomes smaller than propagation delay. Collision free

the transmitter or a specific number of retransmission has been reached.

packet, or implicitly by making use of the broadcast nature of the medium.

propagation delay is large too (about 0.67s/km).

in the performance of digital communications..

is known as Hybrid-ARQ (HARQ) (Kunal et al., 2010).

space and time.

**3. Error correction techniques** 

**3.1 Automatic repeat request** 

receiver. Horizontal channel is characterized by long multipath spreads compared with vertical one. A multipath with varying impulse response tends to be subjected to an Inter Symbol Interference (ISI) that causes severe degradation in the acoustic signals. • High delay and delay variance: underwater acoustic signal speed is just 1500 m/s, which is lower than electromagnetic signal by more than 5 orders of magnitude. The

We organize the rest of this chapter as follows: In section 2, we briefly review the basics of acoustic propagation. In section 3, we review the two most widely used underwater error correction techniques, ARQ and FEC. In section 4, we present mathematical and simulation analysis of the energy efficiency of the techniques. In section 5, an Adaptive Hybrid Energy Efficient Error Correction (AHEC) technique for Underwater Wireless Sensor Networks (UWSN) data transmission is presented. Finally, in section 6, we conclude the chapter.

## **2. Basics of acoustic propagation**

Factors that characterized underwater acoustic propagation include path loss, noise, multipath, Doppler spread, and high and variable propagation delay. Those factors are the main reason for the variability in the acoustic channel. Bandwidth varies from a few KHz in a long range system which operates over several tens of kilometers to more than hundred KHz in a short-range system that operates over several tens of meters. UAC system are classified according to their communication ranges as shown in Table 1 (Akyildiz, et al., 2004, 2005):


Table 1. Underwater Acoustic Communication System Ranges

Below are the factors that characterize underwater acoustic propagation (Colin et al., 2007; Joshy and Babu, 2010; Stojanovic and Preisig, 2009):

	- Attenuation: this is the loss due to the conversion of acoustic energy into heat which known as absorption loss.
	- Spreading: this is the loss due to the expansion of the signal energy over a large area as the wave propagates forward.
	- Man-made noise: it is caused by machinery (pumps, reduction gears, etc), shipping activities, etc.
	- Ambient noise: this is caused by the movement of water which includes tides, current, storms, wind, and rain. It is also caused by biological phenomena. Ambient noise depends mainly on frequency, so it must be considered when selecting frequency band in underwater communications systems (Preisig , 2007).

receiver. Horizontal channel is characterized by long multipath spreads compared with vertical one. A multipath with varying impulse response tends to be subjected to an Inter Symbol Interference (ISI) that causes severe degradation in the acoustic signals.


Most of the factors mentioned above are caused by the chemical-physical properties of the sea water such as temperature, salinity and density, which these factors vary with depth, space and time.

## **3. Error correction techniques**

48 Energy Efficiency in Communications and Networks

We organize the rest of this chapter as follows: In section 2, we briefly review the basics of acoustic propagation. In section 3, we review the two most widely used underwater error correction techniques, ARQ and FEC. In section 4, we present mathematical and simulation analysis of the energy efficiency of the techniques. In section 5, an Adaptive Hybrid Energy Efficient Error Correction (AHEC) technique for Underwater Wireless Sensor Networks (UWSN) data transmission is presented. Finally, in section 6, we conclude the chapter.

Factors that characterized underwater acoustic propagation include path loss, noise, multipath, Doppler spread, and high and variable propagation delay. Those factors are the main reason for the variability in the acoustic channel. Bandwidth varies from a few KHz in a long range system which operates over several tens of kilometers to more than hundred KHz in a short-range system that operates over several tens of meters. UAC system are classified according to their communication ranges as shown in Table 1 (Akyildiz, et al.,

> Very long 1000 <1 Long 10-1000 2-5 Medium 1-10 =10 Short 0.1 -1 20-50 Very short <0.1 >100

Below are the factors that characterize underwater acoustic propagation (Colin et al., 2007;

• Path loss: there are two main sources for path losses for underwater acoustic

• Attenuation: this is the loss due to the conversion of acoustic energy into heat

• Spreading: this is the loss due to the expansion of the signal energy over a large

• Man-made noise: it is caused by machinery (pumps, reduction gears, etc), shipping

• Ambient noise: this is caused by the movement of water which includes tides, current, storms, wind, and rain. It is also caused by biological phenomena. Ambient noise depends mainly on frequency, so it must be considered when selecting

frequency band in underwater communications systems (Preisig , 2007). • Multipath: In most environments, the ocean can be modelled as a wave guide for communication signals. This waveguide is characterized by a reflecting surface and ocean bottom and a variant sound speed. Reflection, refraction and diffraction will occur with those surfaces resulting in multiple propagation paths from the source to the

Table 1. Underwater Acoustic Communication System Ranges

Joshy and Babu, 2010; Stojanovic and Preisig, 2009):

which known as absorption loss.

• Noise: there are two kind of noise:

activities, etc.

area as the wave propagates forward.

Range (Km) Bandwidth (KHz)

**2. Basics of acoustic propagation** 

2004, 2005):

propagation:

Automatic Repeat reQuest (ARQ) and Forward Error Correction (FEC) are two commonly used strategies to combat error in underwater transmission (Bin et al, 2008). ARQ which proposes retransmission (Kunal et al., 2010), are widely used in data communications system for error control as they are simple and provide high reliability. However, the throughput is not constant and decreases rapidly in high bit error rate cases (Lin et al., 1984). In FEC, redundancy is added for error prevention. Redundant bits are encapsulated with data bits to form encoded information. However this increases the payload for transmission. Addition of redundant bit is known as channel coding. Error Correcting Codes (ECC) (block or convolutional) are used for this purpose. FEC codes have constant throughput which is equal to the code rate. However it has the drawback of using parity bit irrespective of the existence of errors. Reliability can be enhanced by combining FEC and ARQ, forming what is known as Hybrid-ARQ (HARQ) (Kunal et al., 2010).

#### **3.1 Automatic repeat request**

ARQ uses error detection codes, acknowledgement and/or negative acknowledgement messages, and time out to retransmit error packet. The basic idea is that the transmitter after sending the packet waits for specific time (time out) to receive an acknowledgement. If it receives positive acknowledgement (ACK), it sends the next packet. On the other hand, if it receives negative acknowledge (NAC) or timed out before receiving any acknowledgement, it then retransmits the same packet. The process repeats until an ACK has been received by the transmitter or a specific number of retransmission has been reached.

In (Tan et al., 2007), an opportunistic (hybrid implicit/explicit) acknowledgement scheme suitable for stop and wait protocols in underwater is proposed. The simple stop and wait (S &W) protocol is chosen as it is the most popular method in underwater acoustic communication due to the half-duplex property of acoustic modem. In the context of a multi hop channel, the work in (Tan et al., 2007) proposed that the acknowledgement can be achieved explicitly by transmitting an acknowledgement packet per successfully received packet, or implicitly by making use of the broadcast nature of the medium.

In (Lee et al., 2008), the channel sharing property inherent in underwater environment is utilized in proposing an efficient ARQ scheme. In this scheme packet size is controlled in such a way that transmission time becomes smaller than propagation delay. Collision free

Energy Efficient Communication for Underwater Wireless Sensors Networks 51

Most telecommunication systems use fixed types of FEC code, which is designed for the expected worst case bit error rate. These codes will fail if the bit error rate ever gets worse. In (Guo, 2006), error recovery through network coding was explored for underwater sensor networks. The computational power of underwater sensors along with the multiple routes provided by the broadcast nature of acoustic medium are the main reasons for applying network coding. In this technique the source and intermediate nodes encode packets and send them on multiple routes. The packets are then recovered in the destination by

In (Xie and Cui, 2007), the Segmented Data Reliable Transport protocol (SDRT) is proposed,. The protocol is a hybrid of FEC and ARQ. It sends data block by block and hop by hop. The sender encodes the packet using erasure codes, and sends it to an intermediate node. The intermediate node reconstructs the packet and encodes it and sends it to the next hop. The sender continues to send the data until it receive an acknowledgement from its next node, and this is the main problem with SDRT as it wastes energy. SDRT however improves

In ( Liu et al., 2010; Bin et al., 2008), the Internode distance-based Redundancy Reliable Transport Protocol (ARRTP) is proposed. It is a hybrid of two types of error correction techniques which encode message on bit and/or packet level. ARRTP is based on distance as adaptation factor. For each range of distance, one or a hybrid of two techniques is used. The technique was also investigated in cooperation mode, making use of the broadcast nature of acoustic signal. ARRTP is found to have better probability of success and energy efficient in single and multi-transmission. This technique is based on fixed channel

conditions analysis, so it is unsuitable in variable underwater channel conditions.

techniques should take into consideration the energy conservation requirement.

and different channel conditions (wind speed and shipping factors) are presented.

**4. Transmission energy efficiency mathematical and simulation analysis** 

Underwater acoustic channel are characterized by variable channel conditions and variable distances between sensor nodes due to water currents. As said earlier in the chapter, in such situations, reliable and efficient communication data transport is needed. Reliability is usually achieved by using error correction techniques. However, energy consumption needs to be considered as it is difficult to recharge or even replace batteries for a large number and sparsely distributed sensors. This condition is even worse in underwater due to the harsh aquatic medium (Colin, 2007; Xie and Cui, 2007). Hence, the design of error correction

In this section, we first develop a model for underwater propagation. A mathematical analysis for energy efficiency for FEC and ARQ techniques in underwater environment is then presented. The analysis is based on communication distance and packet size, and considers the effects of wind speed, and shipping factor. Simulation was done using MATLAB to validate the mathematical analysis results. Results depicting the energy efficiencies of transmission using ARQ and FEC for different packet size, different distances,

The propagation model is responsible for calculating the SNR at the receiver after attenuation and noise are taken into account. To calculate the SNR at the receiver, both the

combining packets from different routes.

**4.1 Underwater propagation model** 

channel utilization and simplify protocol management.

transmission between multiple nodes is achieved by scheduling packets. In a multiple hop setup, the acknowledgement packet is replaced by overhearing packet transmitted from next hop. Overhearing as an acknowledgement method not only saves energy but it also minimizes overhead and transmission latency. The scheme is evaluated by comparing it with an existing stop and wait ARQ in term of the latency, and it shows a reduction in the latency. The latency and energy efficiency is still a problem in bad channel conditions cases though.

In (Gao et al., 2009), the authors make use of the long propagation delay in underwater environment to transmit and receive in a juggling manner. This juggling scheme enables a continuous ARQ to be implemented irrespective of the half-duplex property of the acoustic modem. This scheme decreases the propagation time by having more than one packet in the channel between transmitter and receiver. This leads to high throughput compared with the other variant ARQ schemes, but it is still unsuitable in bad channel conditions or in a longer distance ranges.

In (Valera et al. 2009), a modular and lightweight of an opportunistic multi-hop ARQ (Tan et al., 2007) was implemented for real system. An extensible network stack suitable for challenged underwater acoustic networks was designed and implemented in the work. Evaluation demonstrated that the opportunistic ARQ can provide significant improvement in terms of data delivery ratio. The disadvantage of this technique is an increase in end-toend delay due to queuing and retransmissions.

### **3.2 Forward error correction technique**

Forward Error Correction (FEC) or error control coding is a system for achieving reliable message transmission in a communication system by correcting errors in the receiver side (hence the name 'Forward').

Recent and major activities on error control coding can be summarized as follows:


Forward error correction can be used in two levels, namely at the bit and the byte level. Bit level correction is achieved by adding redundant bits to the data in the sender. At the packet level, additional check packets are transmitted to help recover lost packets. In the FEC, no back channel is needed, but high bandwidth is required. It is therefore suitable in cases where retransmission is costly or impossible, as in broadcasting. The numbers of errors which can be corrected depend on the code rate and the type of coding used. Therefore, different FEC codes are suitable for different conditions.

There are two main types of FEC; the first one is the block codes which work on a fixed-size blocks (packets of bits or symbols), the most famous block codes are Reed-Solemn, Golay, (Bose, Chandhuri and Hocquenghem) BCH code, multidimensional parity and Hamming codes. The other type of FEC is convolutional codes, which work on bit or symbol streams of arbitrary length. It is often decoded using Viterbi algorithm, and it can be turned into block code if desired.

50 Energy Efficiency in Communications and Networks

transmission between multiple nodes is achieved by scheduling packets. In a multiple hop setup, the acknowledgement packet is replaced by overhearing packet transmitted from next hop. Overhearing as an acknowledgement method not only saves energy but it also minimizes overhead and transmission latency. The scheme is evaluated by comparing it with an existing stop and wait ARQ in term of the latency, and it shows a reduction in the latency. The latency and energy efficiency is still a problem in bad channel conditions cases

In (Gao et al., 2009), the authors make use of the long propagation delay in underwater environment to transmit and receive in a juggling manner. This juggling scheme enables a continuous ARQ to be implemented irrespective of the half-duplex property of the acoustic modem. This scheme decreases the propagation time by having more than one packet in the channel between transmitter and receiver. This leads to high throughput compared with the other variant ARQ schemes, but it is still unsuitable in bad channel conditions or in a longer

In (Valera et al. 2009), a modular and lightweight of an opportunistic multi-hop ARQ (Tan et al., 2007) was implemented for real system. An extensible network stack suitable for challenged underwater acoustic networks was designed and implemented in the work. Evaluation demonstrated that the opportunistic ARQ can provide significant improvement in terms of data delivery ratio. The disadvantage of this technique is an increase in end-to-

Forward Error Correction (FEC) or error control coding is a system for achieving reliable message transmission in a communication system by correcting errors in the receiver side

Forward error correction can be used in two levels, namely at the bit and the byte level. Bit level correction is achieved by adding redundant bits to the data in the sender. At the packet level, additional check packets are transmitted to help recover lost packets. In the FEC, no back channel is needed, but high bandwidth is required. It is therefore suitable in cases where retransmission is costly or impossible, as in broadcasting. The numbers of errors which can be corrected depend on the code rate and the type of coding used. Therefore,

There are two main types of FEC; the first one is the block codes which work on a fixed-size blocks (packets of bits or symbols), the most famous block codes are Reed-Solemn, Golay, (Bose, Chandhuri and Hocquenghem) BCH code, multidimensional parity and Hamming codes. The other type of FEC is convolutional codes, which work on bit or symbol streams of arbitrary length. It is often decoded using Viterbi algorithm, and it can be turned into block

Recent and major activities on error control coding can be summarized as follows: • Research on good structural properties, and high error correcting performance.

• Applicability of coding in various transmission system and channels.

though.

distance ranges.

end delay due to queuing and retransmissions.

• Efficient encoding and decoding strategies.

different FEC codes are suitable for different conditions.

**3.2 Forward error correction technique** 

(hence the name 'Forward').

code if desired.

Most telecommunication systems use fixed types of FEC code, which is designed for the expected worst case bit error rate. These codes will fail if the bit error rate ever gets worse.

In (Guo, 2006), error recovery through network coding was explored for underwater sensor networks. The computational power of underwater sensors along with the multiple routes provided by the broadcast nature of acoustic medium are the main reasons for applying network coding. In this technique the source and intermediate nodes encode packets and send them on multiple routes. The packets are then recovered in the destination by combining packets from different routes.

In (Xie and Cui, 2007), the Segmented Data Reliable Transport protocol (SDRT) is proposed,. The protocol is a hybrid of FEC and ARQ. It sends data block by block and hop by hop. The sender encodes the packet using erasure codes, and sends it to an intermediate node. The intermediate node reconstructs the packet and encodes it and sends it to the next hop. The sender continues to send the data until it receive an acknowledgement from its next node, and this is the main problem with SDRT as it wastes energy. SDRT however improves channel utilization and simplify protocol management.

In ( Liu et al., 2010; Bin et al., 2008), the Internode distance-based Redundancy Reliable Transport Protocol (ARRTP) is proposed. It is a hybrid of two types of error correction techniques which encode message on bit and/or packet level. ARRTP is based on distance as adaptation factor. For each range of distance, one or a hybrid of two techniques is used. The technique was also investigated in cooperation mode, making use of the broadcast nature of acoustic signal. ARRTP is found to have better probability of success and energy efficient in single and multi-transmission. This technique is based on fixed channel conditions analysis, so it is unsuitable in variable underwater channel conditions.

## **4. Transmission energy efficiency mathematical and simulation analysis**

Underwater acoustic channel are characterized by variable channel conditions and variable distances between sensor nodes due to water currents. As said earlier in the chapter, in such situations, reliable and efficient communication data transport is needed. Reliability is usually achieved by using error correction techniques. However, energy consumption needs to be considered as it is difficult to recharge or even replace batteries for a large number and sparsely distributed sensors. This condition is even worse in underwater due to the harsh aquatic medium (Colin, 2007; Xie and Cui, 2007). Hence, the design of error correction techniques should take into consideration the energy conservation requirement.

In this section, we first develop a model for underwater propagation. A mathematical analysis for energy efficiency for FEC and ARQ techniques in underwater environment is then presented. The analysis is based on communication distance and packet size, and considers the effects of wind speed, and shipping factor. Simulation was done using MATLAB to validate the mathematical analysis results. Results depicting the energy efficiencies of transmission using ARQ and FEC for different packet size, different distances, and different channel conditions (wind speed and shipping factors) are presented.

#### **4.1 Underwater propagation model**

The propagation model is responsible for calculating the SNR at the receiver after attenuation and noise are taken into account. To calculate the SNR at the receiver, both the

Energy Efficient Communication for Underwater Wireless Sensors Networks 53

It is well known that SNR of an emitted underwater signal at the receiver is given by (Yang

where N(f), A(l,f) are in dBs given from equations (2) and (7). Assuming Omni-directional

2 *Pt <sup>I</sup>* π

The data packet format in ARQ case can be presented as in Figure 1 (a). It consists of a header field α bits long, payload of size *n* bits and a Frame Check Sequence (FCS) τ bits

In FEC case it can be presented as in Figure 1 (b). It consists of a payload of size (n-k) bits

Header FCS Payload α τ *n*

Header Parity check Payload α k n-k

Energy efficiency is the suitable metric which captures both energy and reliability

(1 )

(1 ) *eff*

*PER*

*<sup>E</sup>* = − (10)

*tot*

*<sup>E</sup>* denotes the energy

*E*

*<sup>e</sup>* is the energy throughput, r = (1-PER) is the Packet

constraints, and it is defined as (Sankarasubramaniam et al, 2003; Tian et al. , 2008):

*tot*

*E*

η η= − *<sup>e</sup> PER*

η

Acceptance Rate (PAR), which accounts for data reliability, and *eff*

*SNR SL A l f N f DI* =− − − (, ) ( ) (8)

*<sup>I</sup> SL*

1

*<sup>H</sup>* <sup>=</sup> (9)

μ

*Pa* <sup>=</sup> , where I is the

and B. Liu, 2009; Harris and M. Zorzi, 2007; Brekhovskikh and Lysanov, L.1982)

directivity, directivity index (DI) = 0. The source level 20log

Where Pt is the transmission power, and H is the water depth in m.

intensity at 1 m from the source in watt/m2, given by:

**4.2 Energy efficiency mathematical analysis** 

long. The acknowledgement packet length is ack.

Fig. 1. (a): ARQ Packet Format

Fig. 1. (b): FEC Packet Format

is the energy efficiency,

**4.2.1 Optimization metric** 

Whereη

long, a parity check of k bits and a header field α bits long.

**4.1.3 Signal to noise ratio** 

attenuation of the acoustic signal in water and the ambient noise need to be calculated. The total attenuation is calculated based on the spreading losses and Thorp approximation for the absorption loss (Urick, 1983; Yang and Liu, 2009; Liu et al., 2010; Harris and Zorzi, 2007).

#### **4.1.1 Attenuation**

Attenuation consists of two parts, the first one is the absorption loss and the second part is the spreading loss. To calculate the absorption loss at a given frequency, Thorp's approximation function divides the frequencies into two groups; one group under 400 Hz and the other one over 400 Hz as follows:

$$10\log a(f) = 0.11\frac{f^2}{1+f^2} + 44\frac{f^2}{4200+f} + 2.75 \times 10^{-4} f^2 + 0.003 \text{ f} \approx 0.4$$

$$= 0.002 + 0.11 \times \left(\frac{f}{1+f}\right) + 0.011 f \text{ f} \tag{1}$$

where a(f) is given in dB/km and f in KHz for underwater communications. Combining absorption effects and spreading loss, the total attenuation is as follows:

$$110\log A(l,f) = k\log l + l \times 10\log a(f) \tag{2}$$

where the first term is the spreading loss and the second term is the absorption loss. The spreading coefficient k defines the geometry of the propagation (i.e., k = 1 for cylindrical propagation (shallow water), k = 2 for spherical propagation (deep water), and k = 1.5 for practical spreading) (Urick, 1983).

#### **4.1.2 Noise**

The background noise in ocean has many sources which vary with frequency and location (Wenz and Gordon, 1939). The following formulas give the power spectral density of the four noise components (Yang and Liu, 2009; Liu, 2010; Harris and Zorzi, 2007; Webb, 1992):

$$10\log N\_t(f) = 17 - 30\log(f) \tag{3}$$

$$10\log N\_s(f) = 40 + 20(s - 0.5) + 26\log(f) - 60\log(f + 0.03) \tag{4}$$

$$10\log N\_w(f) = 50 + 7.5 \times w^{0.5} + 20\log(f) - 40\log(f + 0.4) \tag{5}$$

$$10\log N\_{th}(f) = -15 + 20\log(f) \tag{6}$$

Where Nt is the noise due to turbulence, Ns is the noise due to shipping (the shipping variable s take the values between 0 and 1), Nw is the noise due to wind (the wind variables w represent wind speed in m/s), and Nth represents thermal noise. The overall noise power spectral density for a given frequency f (KHz) is then:

$$N(f) = N\_t(f) + N\_s(f) + N\_w(f) + N\_{th}(f) \tag{7}$$

#### **4.1.3 Signal to noise ratio**

52 Energy Efficiency in Communications and Networks

attenuation of the acoustic signal in water and the ambient noise need to be calculated. The total attenuation is calculated based on the spreading losses and Thorp approximation for the absorption loss (Urick, 1983; Yang and Liu, 2009; Liu et al., 2010; Harris and Zorzi, 2007).

Attenuation consists of two parts, the first one is the absorption loss and the second part is the spreading loss. To calculate the absorption loss at a given frequency, Thorp's approximation function divides the frequencies into two groups; one group under 400 Hz

4 2

10log *Al f k l l a f* ( , ) lo = +× g 10log ( ) (2)

10log ( ) 17 30log( ) *N f <sup>t</sup>* = − *f* (3)

10log ( ) 15 20log( ) *N f th* =− + *f* (6)

() () () () () *Nf N f N f N f N f* =++ + *t s w th* (7)

10log ( ) 40 20( 0.5) 26log( ) 60log( 0.03) *Nf s <sup>s</sup>* =+ − + − + *f f* (4)

0.5 10log ( ) 50 7.5 20log( ) 40log( 0.4) *Nf w f f <sup>w</sup>* =+× + − + (5)

f>0.4

f< 0.4 (1)

2 2

+ +

= +× +

absorption effects and spreading loss, the total attenuation is as follows:

<sup>2</sup> 10log ( ) 0.11 44 2.75 10 0.003 <sup>1</sup> <sup>4200</sup> *f f a f <sup>f</sup> <sup>f</sup> <sup>f</sup>* <sup>−</sup> = + +× +

0.002 0.11 ( ) 0.011 <sup>1</sup>

+

where a(f) is given in dB/km and f in KHz for underwater communications. Combining

where the first term is the spreading loss and the second term is the absorption loss. The spreading coefficient k defines the geometry of the propagation (i.e., k = 1 for cylindrical propagation (shallow water), k = 2 for spherical propagation (deep water), and k = 1.5 for

The background noise in ocean has many sources which vary with frequency and location (Wenz and Gordon, 1939). The following formulas give the power spectral density of the four noise components (Yang and Liu, 2009; Liu, 2010; Harris and Zorzi, 2007; Webb, 1992):

Where Nt is the noise due to turbulence, Ns is the noise due to shipping (the shipping variable s take the values between 0 and 1), Nw is the noise due to wind (the wind variables w represent wind speed in m/s), and Nth represents thermal noise. The overall noise power

*<sup>f</sup> <sup>f</sup> <sup>f</sup>*

**4.1.1 Attenuation** 

and the other one over 400 Hz as follows:

practical spreading) (Urick, 1983).

spectral density for a given frequency f (KHz) is then:

**4.1.2 Noise** 

It is well known that SNR of an emitted underwater signal at the receiver is given by (Yang and B. Liu, 2009; Harris and M. Zorzi, 2007; Brekhovskikh and Lysanov, L.1982)

$$SNR = SL - A(l, f) - N(f) - DI \tag{8}$$

where N(f), A(l,f) are in dBs given from equations (2) and (7). Assuming Omni-directional directivity, directivity index (DI) = 0. The source level 20log 1 *<sup>I</sup> SL* μ*Pa* <sup>=</sup> , where I is the intensity at 1 m from the source in watt/m2, given by:

$$I = \frac{P\_t}{2\pi H} \tag{9}$$

Where Pt is the transmission power, and H is the water depth in m.

#### **4.2 Energy efficiency mathematical analysis**

The data packet format in ARQ case can be presented as in Figure 1 (a). It consists of a header field α bits long, payload of size *n* bits and a Frame Check Sequence (FCS) τ bits long. The acknowledgement packet length is ack.

In FEC case it can be presented as in Figure 1 (b). It consists of a payload of size (n-k) bits long, a parity check of k bits and a header field α bits long.


Fig. 1. (a): ARQ Packet Format


Fig. 1. (b): FEC Packet Format

#### **4.2.1 Optimization metric**

Energy efficiency is the suitable metric which captures both energy and reliability constraints, and it is defined as (Sankarasubramaniam et al, 2003; Tian et al. , 2008):

$$
\eta = \eta\_c (1 - PER)
$$

$$
= \frac{E\_{eff}}{E\_{tot}} (1 - PER) \tag{10}
$$

Whereη is the energy efficiency, η *<sup>e</sup>* is the energy throughput, r = (1-PER) is the Packet Acceptance Rate (PAR), which accounts for data reliability, and *eff tot E <sup>E</sup>* denotes the energy

Energy Efficient Communication for Underwater Wireless Sensors Networks 55

assuming independent bit errors, the Packer Error Rate (PER) for ARQ can be derived as

From equation (9) energy efficiency of ARQ without retransmission strategy can hence be

( ) (1 ) ( )( ) *tr re tr ARQ*

> (1 ) ( ) *ARQ <sup>n</sup> PER*

where *eff EARQ* is the energy consumed by the payload only, *tot EARQ* is the total energy consumed.

Using convolution turbo code as forward error correction techniques, encoding (Eenc) and decoding energy (Edec) are considered to be negligibly small (Sankarasubramaniam et al., 2003; Tian et al., 2008), and from Figure 1 (b), the expression for the energy efficiency is

> *eff FEC FEC tot FEC FEC <sup>E</sup> Eff PER E* = −

*tr re tr*

+ − = − + +

*P Pn T*

*n* α

( )( ) (1 ) ( )( ) *tr re tr FEC*

> ( ) (1 ) ( ) *FEC n k PER*

Simulation offers a powerful tool to validate mathematical analysis. The simulation is carried out for a two system using different error correction techniques using MATLAB.

*P P n kT PER*

α

(1 )

*P P nT PER*

+ + α τ

(1 )

= − + ++ (18)

*tr re E E E EE FEC FEC FEC dec enc* = + ++ (19)

<sup>−</sup> = − + (20)

=− − (17)

1 (1 )*<sup>n</sup> PERARQ Pb*

*eff ARQ ARQ tot ARQ ARQ*

*E Eff PER <sup>E</sup>* = −

*tr re tr*

<sup>+</sup> = − + + ++

*P P n ack T* α τ

> *n ack* α τ

**4.2.4 FEC energy efficiency mathematical derivation** 

The energy consumption of FEC is given by:

where *PERFEC* is calculated using equation (13).

This expression closely approximates PER under bursty error conditions.

follows:

written as:

defined as:

**4.3 Simulation** 

throughput. Therefore, the energy efficiency η represents the useful fraction of the total energy expenditure in a communication link between sensors.

#### **4.2.2 Bit error rate calculation**

Using 8-Phase Shift Keying (PSK) scheme as the suitable modulation techniques for underwater acoustic communication, the symbol error probability Ps for ARQ is given by (Labrador et al., 2009):

$$P\_s = 2Q(\sqrt{2\gamma\_s}\sin\frac{\pi}{M})\tag{11}$$

where M=8 for 8-PSK, and the bit error probability Pb is given by:

$$P\_b = \bigvee\_{\mathfrak{B}}^{P\_s} \tag{12}$$

Whereas for FEC convolution code (Lee et al., 2008):

$$P\_b = \frac{1}{k} \sum\_{d=d\_{fm}}^{\circ} w(d) Q(\sqrt{2d \mathcal{R}\_c \mathcal{Y}\_b} \tag{13}$$

where w(d) is the weight distribution function, dfree is the minimum hamming distance, and *<sup>b</sup>* γ is the received SNR, <sup>1</sup> *<sup>c</sup> <sup>R</sup> <sup>k</sup> <sup>k</sup>* <sup>=</sup> <sup>+</sup> is the code rate.

#### **4.2.3 ARQ energy efficiency mathematical analysis**

Energy consumption of sensor node for communication in one hop is given by:

$$E\_{ARQ} = E\_{ARQ}^{tr} + E\_{ARQ}^{tr} \tag{14}$$

Where *tr EARQ* is the energy consumed by the sender in transmitting the data and receiving the acknowledgement, and *re EARQ* is the energy consumed by the receiver in receiving the data and transmitting the acknowledgement as presented in the following equations:

$$\begin{aligned} \mathbf{E}\_{ARQ}^{tr} &= \mathbf{E}\_{data}^{tr} + \mathbf{E}\_{ack}^{re} \\\\ \mathbf{E} &= P\_{tr} \mathbf{I}\_{data} \mathbf{T}\_{tr} + P\_{re} \mathbf{I}\_{ack} \mathbf{T}\_{tr} \\\\ \mathbf{E}\_{ARQ}^{re} &= \mathbf{E}\_{data}^{re} + \mathbf{E}\_{ack}^{tr} \\\\ \mathbf{E} &= P\_{re} \mathbf{I}\_{data} \mathbf{T}\_{tr} + P\_{tr} \mathbf{I}\_{ack} \mathbf{T}\_{tr} \end{aligned} \tag{16}$$

Where *Ptr re* / is the power consumed in transmitting/ receiving, and <sup>1</sup> *Ttr <sup>R</sup>* <sup>=</sup> is the time of transmitting 1 bit. From Figure 1 (a), using the bit error rate probability Pb in (12), and assuming independent bit errors, the Packer Error Rate (PER) for ARQ can be derived as follows:

$$PER\_{ARQ} = 1 - (1 - P\_b)^{n + \alpha + r} \tag{17}$$

This expression closely approximates PER under bursty error conditions.

From equation (9) energy efficiency of ARQ without retransmission strategy can hence be written as:

$$E\mathcal{J}\_{ARQ} = \frac{E\_{ARQ}^{\text{eff}}}{E\_{ARQ}^{\text{tot}}} (1 - PER\_{ARQ})$$

$$= \frac{(P\_{tr} + P\_{rc})nT\_{tr}}{(P\_{tr} + P\_{rc})(n + \alpha + \tau + ack)T\_{tr}} (1 - PER\_{ARQ})$$

$$= \frac{n}{(n + \alpha + \tau + ack)} (1 - PER\_{ARQ})\tag{18}$$

where *eff EARQ* is the energy consumed by the payload only, *tot EARQ* is the total energy consumed.

#### **4.2.4 FEC energy efficiency mathematical derivation**

The energy consumption of FEC is given by:

$$E\_{\rm FEC} = E\_{\rm FEC}^{tr} + E\_{\rm FEC}^{re} + E\_{\rm dec} + E\_{\rm enc} \tag{19}$$

Using convolution turbo code as forward error correction techniques, encoding (Eenc) and decoding energy (Edec) are considered to be negligibly small (Sankarasubramaniam et al., 2003; Tian et al., 2008), and from Figure 1 (b), the expression for the energy efficiency is defined as:

$$Ef\_{FEC}^{\prime} = \frac{E\_{FEC}^{\prime \prime}}{E\_{FEC}^{tot}} (1 - PER\_{FEC})$$

$$= \frac{(P\_{tr} + P\_{re})(n - k)T\_{tr}}{(P\_{tr} + P\_{re})(n + \alpha)T\_{tr}} (1 - PER\_{FEC})$$

$$= \frac{(n - k)}{(n + \alpha)} (1 - PER\_{FEC}) \tag{20}$$

where *PERFEC* is calculated using equation (13).

#### **4.3 Simulation**

54 Energy Efficiency in Communications and Networks

η

Using 8-Phase Shift Keying (PSK) scheme as the suitable modulation techniques for underwater acoustic communication, the symbol error probability Ps for ARQ is given by

*P Q s s* 2 ( 2 sin *<sup>M</sup>*

<sup>1</sup> ()( 2

where w(d) is the weight distribution function, dfree is the minimum hamming distance,

Where *tr EARQ* is the energy consumed by the sender in transmitting the data and receiving the acknowledgement, and *re EARQ* is the energy consumed by the receiver in receiving the data

*tr tr re E EE ARQ data ack* = +

*re re tr E EE ARQ data ack* = +

Where *Ptr re* / is the power consumed in transmitting/ receiving, and <sup>1</sup> *Ttr <sup>R</sup>* <sup>=</sup> is the time of transmitting 1 bit. From Figure 1 (a), using the bit error rate probability Pb in (12), and

*free b c b*

*<sup>k</sup>* <sup>=</sup> <sup>+</sup> is the code rate.

*d d P w d Q dR <sup>k</sup>*

Energy consumption of sensor node for communication in one hop is given by:

and transmitting the acknowledgement as presented in the following equations:

=

∞

 ≈ γ

> 3 *<sup>s</sup> <sup>b</sup>*

π

γ

represents the useful fraction of the total

(11)

*<sup>P</sup> <sup>P</sup>* <sup>=</sup> (12)

<sup>=</sup> (13)

*tr re EEE ARQ ARQ ARQ* = + (14)

= + *Pl T Pl T tr data tr re ack tr* (15)

= + *Pl T Pl T re data tr tr ack tr* (16)

throughput. Therefore, the energy efficiency

**4.2.2 Bit error rate calculation** 

(Labrador et al., 2009):

and *<sup>b</sup>* γ

energy expenditure in a communication link between sensors.

where M=8 for 8-PSK, and the bit error probability Pb is given by:

Whereas for FEC convolution code (Lee et al., 2008):

**4.2.3 ARQ energy efficiency mathematical analysis** 

is the received SNR, <sup>1</sup> *<sup>c</sup> <sup>R</sup> <sup>k</sup>*

Simulation offers a powerful tool to validate mathematical analysis. The simulation is carried out for a two system using different error correction techniques using MATLAB.

Energy Efficient Communication for Underwater Wireless Sensors Networks 57

In the receiver side 8-PSK demodulator is used to demodulate the signal using the function:

then BER is obtained by comparing the input x with the output z using the function:

In the second system the 8-PSK is replaced by a convolutional encoder with the following:

The value of SNR in underwater channel is written as in section 2.1.3., and an AWGN

In the receiver side, the punctured code is decoded by viterbi using the function:

The energy efficiency is calculated from the BER as in section 2.2.4

(LinQuest Inc., 2011), and the parameters as given in Table 3:


then BER is obtained by comparing the input x with the output decoded using the function:

The results are obtained using a MATLAB, assuming LinkQuest UWM2000 acoustic modem

First, a suitable frequency range based on AN Factor as in Figure 2 was calculated; this frequency range corresponds to the minimum AN factor. A suitable range is found from 10

From Figures 3 (a) and 3 (b), it is clear that transmission energy efficiency of both techniques increases with increasing packet size in short distances, whereas decreases in long distances

KHz up to 25 KHz, below and above this range the AN Factor increases sharply.





**4.4 Results and analysis** 

function is used as:

This puncturing is for 5/6 code rate.





Then puncturing is attained by the following function:

Then 0 bit is mapped to 1 and 1 bit to -1 using the function:


Trellis is defined using the following function:

The energy efficiency is calculated from the BER as in section 2.2.3.

Two types of parameters are considered for design and configuration. Energy efficiency and packet probability of success (PAR) are taken as the main performance factors in comparing between the two systems.

#### **4.3.1 Design parameters**

The design parameters are the parameters that can be varied in order to study their effect on the system energy efficiency. In the first system ARQ technique is used as the error correction technique, where 8-PSK is used as the modulation technique as it is the best modulation technique in underwater channel as stated in the literature. In the second convolutional coding is used as the FEC error correction technique (Labrador et al., 2009).

The design parameters used are the distance, shipping factor and wind speed. Shipping factor and wind speed are taken as a representative for variable channel conditions; any other channel condition factor will have the same effect.

Modulation and encoding technique types and design parameters can be written as in Table 2. below:


Table 2. Modulation, Encoding Types and Design Parameters

#### **4.3.2 Configuration parameters**

The simulation is carried out using MATLAB.

In the transmitter a random bit generator is used with the parameters as follows:


Binary data stream are created as a column vector using the function: x = randint (n, 1);

A Bit-to-Symbol mapping which convert the bits in x into k-bit symbols is done using the following MATLAB function:


Then an 8-PSK modulator is used to modulate the signal with the function:


The value of SNR in underwater channel is calculated as in section 2.1.3., and an AWGN function is used as:


In the receiver side 8-PSK demodulator is used to demodulate the signal using the function:


the Symbol-to-Bit mapping is done using the function:


56 Energy Efficiency in Communications and Networks

Two types of parameters are considered for design and configuration. Energy efficiency and packet probability of success (PAR) are taken as the main performance factors in comparing

The design parameters are the parameters that can be varied in order to study their effect on the system energy efficiency. In the first system ARQ technique is used as the error correction technique, where 8-PSK is used as the modulation technique as it is the best modulation technique in underwater channel as stated in the literature. In the second convolutional coding is used as the FEC error correction technique (Labrador et al., 2009). The design parameters used are the distance, shipping factor and wind speed. Shipping factor and wind speed are taken as a representative for variable channel conditions; any

Modulation and encoding technique types and design parameters can be written as in Table

Parameter Description Type or Value Modulation Modulation technique used in ARQ case 8-PSK

Shipping factor Factor describe the effect of shipping From 0 to 1 Wind speed Factor describe the effect of wind Any value in m/s

In the transmitter a random bit generator is used with the parameters as follows:

Binary data stream are created as a column vector using the function: x = randint (n, 1);

A Bit-to-Symbol mapping which convert the bits in x into k-bit symbols is done using the

The value of SNR in underwater channel is calculated as in section 2.1.3., and an AWGN

Encoding Encoding technique used for error correction Convolution coding Distance Communication distance From 800 to 3000 m

between the two systems.

**4.3.1 Design parameters** 

2. below:

other channel condition factor will have the same effect.

Table 2. Modulation, Encoding Types and Design Parameters


Then an 8-PSK modulator is used to modulate the signal with the function:

**4.3.2 Configuration parameters** 

• size of signal constellation M = 8; • Number of bit per symbol k = 3, • Number of bit processed n = 3e4.

following MATLAB function:

function is used as:

The simulation is carried out using MATLAB.


then BER is obtained by comparing the input x with the output z using the function:


The energy efficiency is calculated from the BER as in section 2.2.3.

In the second system the 8-PSK is replaced by a convolutional encoder with the following:

Trellis is defined using the following function:


Then puncturing is attained by the following function:


This puncturing is for 5/6 code rate.

Then 0 bit is mapped to 1 and 1 bit to -1 using the function:


The value of SNR in underwater channel is written as in section 2.1.3., and an AWGN function is used as:

```
```
In the receiver side, the punctured code is decoded by viterbi using the function:


then BER is obtained by comparing the input x with the output decoded using the function:


The energy efficiency is calculated from the BER as in section 2.2.4

#### **4.4 Results and analysis**

The results are obtained using a MATLAB, assuming LinkQuest UWM2000 acoustic modem (LinQuest Inc., 2011), and the parameters as given in Table 3:

First, a suitable frequency range based on AN Factor as in Figure 2 was calculated; this frequency range corresponds to the minimum AN factor. A suitable range is found from 10 KHz up to 25 KHz, below and above this range the AN Factor increases sharply.

From Figures 3 (a) and 3 (b), it is clear that transmission energy efficiency of both techniques increases with increasing packet size in short distances, whereas decreases in long distances

Energy Efficient Communication for Underwater Wireless Sensors Networks 59

FEC Energy Eff. (w = 0 m/s, s = 0)

<sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>0</sup>

Distance (m)

In Figure 4 (a) transmission energy efficiency of ARQ and FEC for a packet length of 512 bit is shown. It is apparent that transmission using ARQ is more energy efficient than using FEC below a specific distance (cut-off distance), and transmission using FEC is more energy efficient after this distance. The effect of shipping is unseen and can be neglected. In Figure 4 (b) the effect of wind is very clear, and the cut-off distance decreases from 1700 m when no wind exists to 1250 m when the wind speed is 1 m/s. ARQ efficiency starts to decrease at 1600 m when no wind exists, and at 1100 m when the wind speed is 1 m/s, whereas for FEC it starts to decrease at 2500 m when no wind exist and at 1800 m when the wind speed is 1

ARQ Vs FEC (n = 512 bit, w = 0 m/s)

<sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>0</sup>

Distance (m)

Fig. 4. (a): ARQ Vs FEC Transmission Energy Efficiency (n = 512 bit, Variable Shipping

Fig. 3. (b): FEC Transmission Energy Efficiency (Mathematical and Simulation Results)

0.1

0.1

0.2

FEC (s = 1) ARQ (s = 1) FEC (s = 0) ARQ (s = 0)

0.3

0.4

Energy Eff.

0.5

0.6

0.7

0.8

m/s.

Factor)

0.2

n = 512 bit (Simulation) n = 512 bit (Mathematical) n = 1024 bit (Simulation) n = 1024 bit (Mathematic) n = 2048 bit (Simulation) n = 2048 bit (Mathematic)

0.3

0.4

Energy eff.

0.5

0.6

0.7

0.8


Table 3. Simulation Parameters

Fig. 2. AN Factor

for both techniques. It is also clear that there is only a slight differences between mathematical and simulation results which validate the results. This differences between mathematical and simulation results decreases as the number of bits transmitted in the simulation increases.

Fig. 3. (a): ARQ Transmission Energy Efficiency (Mathematical and Simulation Results)

58 Energy Efficiency in Communications and Networks

*Pt* Transmitting Power 2 W *Pre* Receiving Power 0.75 *R* Bit Data Rate 10 kbps *ack l* Acknowledge packet length 7 Byte α + τ Header + FCS length 11 Byte

AN factor

<sup>0</sup> <sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>20</sup>

Distance (m)

for both techniques. It is also clear that there is only a slight differences between mathematical and simulation results which validate the results. This differences between mathematical and simulation results decreases as the number of bits transmitted in the

ARQ Energy Eff. (w =0 m/s, s = 0)

<sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>0</sup>

Distance (m)

Fig. 3. (a): ARQ Transmission Energy Efficiency (Mathematical and Simulation Results)

Definition Quantity

f = 35 Khz f = 30 Khz f = 25 Khz f = 20 Khz f = 15 Khz f = 10 Khz f = 5 Khz f = 1 Khz

Symbol Parameters

Table 3. Simulation Parameters

40

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

n = 512 bit (Simulation) n = 512 bit (Mathematical) n = 1024 bit (Simulation) n = 1024 bit (Mathematical) n = 2048 bit (Simulation) n = 2048 bit (Mathematical)

Energy Eff.

Fig. 2. AN Factor

simulation increases.

60

80

AN factor (dB)

100

120

140

Fig. 3. (b): FEC Transmission Energy Efficiency (Mathematical and Simulation Results)

In Figure 4 (a) transmission energy efficiency of ARQ and FEC for a packet length of 512 bit is shown. It is apparent that transmission using ARQ is more energy efficient than using FEC below a specific distance (cut-off distance), and transmission using FEC is more energy efficient after this distance. The effect of shipping is unseen and can be neglected. In Figure 4 (b) the effect of wind is very clear, and the cut-off distance decreases from 1700 m when no wind exists to 1250 m when the wind speed is 1 m/s. ARQ efficiency starts to decrease at 1600 m when no wind exists, and at 1100 m when the wind speed is 1 m/s, whereas for FEC it starts to decrease at 2500 m when no wind exist and at 1800 m when the wind speed is 1 m/s.

Fig. 4. (a): ARQ Vs FEC Transmission Energy Efficiency (n = 512 bit, Variable Shipping Factor)

Energy Efficient Communication for Underwater Wireless Sensors Networks 61

adaptation algorithm is based on the current Bit Error Rate (BER), current error correction technique, and a pre-calculated Packet Acceptance Rate (PAR) ranges look-up table which is pre-calculated using the energy efficiency derivation has been done in the previous section. Based on this, a periodical 3-bit feedback is added to the acknowledgement packet to tell the sender which error correction technique is most suitable for current channel conditions and distance. The error correction is chosen from a pure ARQ in a good channel conditions and short distances to a hybrid of ARQ and FEC with variable encoding rates in bad channel

This section is organized as follows: the in adaptive hybrid error correction technique is presented in section 5.1. In section 5.2 we show how the pre-calculated PAR ranges look-up table is calculated. Then in section 5.3, we compare the proposed AHEC technique with the techniques that use only ARQ or only FEC as the error correction technique in variable

The results of the derivations in the previous section state that transmission energy efficiency varies with the variation in transmission distances and channel conditions for both the ARQ and the FEC. Depending on the underwater network condition and the internode distances, one technique will be better than the other. The propose AHEC technique is designed to achieve high transmission energy efficiency in such conditions, by

The technique works like this: for variable distances and variable channel conditions, AHEC technique always search for the technique with the highest energy efficiency, and since reliability is one part in transmission energy efficiency calculation as stated in equation (10), it will also be a reliable technique. The technique depends on an adaptation algorithm which based on the current PAR, current encoding technique used, and a pre-calculated PAR ranges look-up table to determine which error correction technique is most suitable for the current distance and current channel conditions. AHEC technique can designed as in the

In AHEC technique, only modulation technique (i.e. ARQ) is used in good channel conditions and short distances, which means low BER. In bad channel conditions and long

Variable code rates are obtained using puncturing technique by deleting a part of the bits of low-rate convolution code (Begin et al., 1990), as in Table 4., and it is represented in

distances a hybrid of ARQ and variable code rates convolutional encoding are used.

Code rate Puncturing Matrix

condition and over longer distance ranges.

channel conditions and over variable distance ranges.

adaptively changing the error correction technique used.

diagram Figure 5.

2/3 [1 1 0 1] 3/4 [1 1 0 1 1 0] 4/5 [1 1 0 1 1 0 1 0] 5/6 [1 1 0 1 1 0 1 0 1 0] 6/7 [1 1 0 1 1 0 1 0 1 0 1 0]

Table 4. Puncturing Matrix

**5.1 Adaptive hybrid error correction technique main concepts** 

Fig. 4. (b): ARQ Vs FEC Transmission Energy Efficiency (n = 512 bit, Variable Wind Speed)

#### **4.5 Discussion**

A mathematical analysis for the energy efficiencies of ARQ and FEC data transmission has been presented. Simulation results validate the mathematical derivation results. It is found that transmission energy efficiency in underwater environment increases with increasing packet size in short distances and decreases with packet size in longer distances. It is also found that transmission using ARQ is more energy efficient below a specific distance (cutoff distance), whereas transmission using FEC is more efficient after that distance. This cutoff distance is affected by wind speed. Shipping factor has been found to have no effect on this frequency values. From those results we can say that variable distances and variable channel conditions which characterize underwater channel make it energy inefficient to use one or fixed type of error correction techniques in transmission.

The results obtained from this part will be the basis for designing and implementing a new adaptive hybrid energy efficient error correction protocol for underwater wireless sensor networks in the next part.

#### **5. Adaptive hybrid energy efficient error correction technique for UWSN**

As it is energy inefficient in transmission to use one or fixed type of error correction in realistic underwater conditions, it is important to consider hybrid error correction technique. This hybrid error correction technique must adapt to the variation in channel conditions and to the variation in distances between sensor nodes.

In this section, we propose an Adaptive Hybrid Energy Efficient Error Correction (AHEC) technique for Underwater Wireless Sensor Networks (UWSN) data transmission. The proposed technique depends on an adaptation algorithm which determines the most energy efficient error correction technique for the current channel conditions and distance. The 60 Energy Efficiency in Communications and Networks

ARQ Vs FEC Energy Eff. (n = 512, s = 0)

<sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>0</sup>

Distace (m)

Fig. 4. (b): ARQ Vs FEC Transmission Energy Efficiency (n = 512 bit, Variable Wind Speed)

A mathematical analysis for the energy efficiencies of ARQ and FEC data transmission has been presented. Simulation results validate the mathematical derivation results. It is found that transmission energy efficiency in underwater environment increases with increasing packet size in short distances and decreases with packet size in longer distances. It is also found that transmission using ARQ is more energy efficient below a specific distance (cutoff distance), whereas transmission using FEC is more efficient after that distance. This cutoff distance is affected by wind speed. Shipping factor has been found to have no effect on this frequency values. From those results we can say that variable distances and variable channel conditions which characterize underwater channel make it energy inefficient to use

The results obtained from this part will be the basis for designing and implementing a new adaptive hybrid energy efficient error correction protocol for underwater wireless sensor

As it is energy inefficient in transmission to use one or fixed type of error correction in realistic underwater conditions, it is important to consider hybrid error correction technique. This hybrid error correction technique must adapt to the variation in channel

In this section, we propose an Adaptive Hybrid Energy Efficient Error Correction (AHEC) technique for Underwater Wireless Sensor Networks (UWSN) data transmission. The proposed technique depends on an adaptation algorithm which determines the most energy efficient error correction technique for the current channel conditions and distance. The

**5. Adaptive hybrid energy efficient error correction technique for UWSN** 

0.1

**4.5 Discussion** 

networks in the next part.

0.2

FEC (w = 1 m/s) ARQ (w = 1 m/s) FEC (w = 0 m/s) ARQ (w = 0 m/s)

one or fixed type of error correction techniques in transmission.

conditions and to the variation in distances between sensor nodes.

0.3

0.4

Energy Eff.

0.5

0.6

0.7

0.8

adaptation algorithm is based on the current Bit Error Rate (BER), current error correction technique, and a pre-calculated Packet Acceptance Rate (PAR) ranges look-up table which is pre-calculated using the energy efficiency derivation has been done in the previous section. Based on this, a periodical 3-bit feedback is added to the acknowledgement packet to tell the sender which error correction technique is most suitable for current channel conditions and distance. The error correction is chosen from a pure ARQ in a good channel conditions and short distances to a hybrid of ARQ and FEC with variable encoding rates in bad channel condition and over longer distance ranges.

This section is organized as follows: the in adaptive hybrid error correction technique is presented in section 5.1. In section 5.2 we show how the pre-calculated PAR ranges look-up table is calculated. Then in section 5.3, we compare the proposed AHEC technique with the techniques that use only ARQ or only FEC as the error correction technique in variable channel conditions and over variable distance ranges.

## **5.1 Adaptive hybrid error correction technique main concepts**

The results of the derivations in the previous section state that transmission energy efficiency varies with the variation in transmission distances and channel conditions for both the ARQ and the FEC. Depending on the underwater network condition and the internode distances, one technique will be better than the other. The propose AHEC technique is designed to achieve high transmission energy efficiency in such conditions, by adaptively changing the error correction technique used.

The technique works like this: for variable distances and variable channel conditions, AHEC technique always search for the technique with the highest energy efficiency, and since reliability is one part in transmission energy efficiency calculation as stated in equation (10), it will also be a reliable technique. The technique depends on an adaptation algorithm which based on the current PAR, current encoding technique used, and a pre-calculated PAR ranges look-up table to determine which error correction technique is most suitable for the current distance and current channel conditions. AHEC technique can designed as in the diagram Figure 5.

In AHEC technique, only modulation technique (i.e. ARQ) is used in good channel conditions and short distances, which means low BER. In bad channel conditions and long distances a hybrid of ARQ and variable code rates convolutional encoding are used.

Variable code rates are obtained using puncturing technique by deleting a part of the bits of low-rate convolution code (Begin et al., 1990), as in Table 4., and it is represented in


Table 4. Puncturing Matrix

Energy Efficient Communication for Underwater Wireless Sensors Networks 63

Using error detection technique in the receiver, BER is periodically calculated, and from the

where J is the suitable error correction technique required, PAR is the current packet acceptance rate, I is the current error correction technique used, and PARMAX(I,J),

> *<sup>n</sup>* ( ) *Ai <sup>n</sup> J n I PAR*

where *<sup>n</sup> Ai* is a look-up table taken from the energy efficiency derivation of six error

1.... .

 <sup>∈</sup> <sup>=</sup> 

From the value of J obtained, a 3-bit feedback is added to the acknowledgement to state

Correction Technique Consists of FEC Code Rate Feedback

1 Pure ARQ 000

2 Hybrid ARQ& FEC 6/7 001

3 Hybrid ARQ& FEC 5/6 010

4 Hybrid ARQ& FEC 4/5 011

5 Hybrid ARQ& FEC 3/4 100

6 Hybrid ARQ& FEC 2/3 101

*otherwise*

Then the suitable error correction technique is calculated from the function:

We can mathematically model this function as in the following formula:

correction techniques (One ARQ and five varying code rates FEC), and

which error correction technique to use as in Table 6. below:

Table 6. Error Correction Techniques Details

6

1

( ) 0..... *<sup>B</sup> if x B I x*

=

(1 )*<sup>n</sup> PAR BER* = − (21)

= × (23)

(24)

*J f PAR I PARMAX I J PARMIN I J* = ( , , ( , ), ( , )) (22)

**5.1.1 AHEC technique adaptation algorithm** 

BER, PAR is calculated using the packet length n as:

PARMIN(I,J) are the pre-calculated PAR ranges look-up.

The adaptation algorithm is as follows:

Fig. 5. AHEC Technique Design

MATLAB using systematic puncturing convolution codes with the parameters obtained from (Begin et al., 1990) as shown in Table 5.


Table 5. Minimum Hamming Distances (dfree) and Weight Distribution (wdfree) for Variable Rate Convolutional Codes.

#### **5.1.1 AHEC technique adaptation algorithm**

The adaptation algorithm is as follows:

62 Energy Efficiency in Communications and Networks

Symbol-To-Bit mapping

Pre-Calculated PAR ranges Look-up table

MATLAB using systematic puncturing convolution codes with the parameters obtained

Bit-To-Symbol mapping

**Demodulation** 

Convolutional Encoding Viterbi Decoding Depuncturing

Modulation

Convolutional Encoding Trellis definition Puncturing

8-PSK Modulation

8-PSK Demodulation Underwater Channel Model

Rc 2/3 3/4 4/5 5/6 6/7 dfree 33 2 2 2

Wdfree 1 15 1 2 5 Wdfree+1 10 104 36 111 186 Wdfree+2 54 540 309 974 1942 Wdfree+3 226 2520 2058 6815 16428 Wdfree+4 853 11048 12031 43598 124469 Wdfree+5 3038 46516 65754 263671 887512 Wdfree+6 10432 190448 344656 1536563 6088910 Wdfree+7 34836 763944 1755310 8724988 40664781 Wdfree+8 114197 3016844 8754128 46801477 266250132 Table 5. Minimum Hamming Distances (dfree) and Weight Distribution (wdfree) for

Fig. 5. AHEC Technique Design

Output data stream

Adaptation Algorithm (3 bits feedback)

Technique Selector

Random binary data stream

> (BER) Using Error detection technique (CRC)

Variable Rate Convolutional Codes.

from (Begin et al., 1990) as shown in Table 5.

Using error detection technique in the receiver, BER is periodically calculated, and from the BER, PAR is calculated using the packet length n as:

$$PAR = (1 - BER)^n \tag{21}$$

Then the suitable error correction technique is calculated from the function:

$$I = f(PAR, I, PARMAX(I, I))\_{\prime} \\ PARMIN(I, I))\tag{22}$$

where J is the suitable error correction technique required, PAR is the current packet acceptance rate, I is the current error correction technique used, and PARMAX(I,J), PARMIN(I,J) are the pre-calculated PAR ranges look-up.

We can mathematically model this function as in the following formula:

$$J = \sum\_{n=1}^{6} n \times I\_{A\_i^v} \text{(PAR)} \tag{23}$$

where *<sup>n</sup> Ai* is a look-up table taken from the energy efficiency derivation of six error correction techniques (One ARQ and five varying code rates FEC), and

$$I\_B(\mathbf{x}) = \begin{cases} 1...\text{if } \mathbf{x} \in B\\ 0...\text{otherwise} \end{cases} \tag{24}$$

From the value of J obtained, a 3-bit feedback is added to the acknowledgement to state which error correction technique to use as in Table 6. below:


Table 6. Error Correction Techniques Details

Energy Efficient Communication for Underwater Wireless Sensors Networks 65

technique number 4.

range, then:

range, then:

**5.3 Results and discussion** 

is the most energy efficient technique.

is the most energy efficient technique.

variable channel conditions and variable distances.

Table 7. Pre-Calculated Look-Up PAR Ranges Table

1 0.95 -1.0 0.95 – 0.0

will be the minimum values in the ranges which makes the suitable technique is

This means the PAR of any technique J at this point = PARMINJ,4 i.e. if the PAR of the current technique J is in between PARMINJ,4 and PARMAXJ,4, then technique number 4

As the minimum values in the fourth ranges equal the maximum values in the fifth

As the minimum values in the fifth ranges equal the maximum values in the sixth

PARMAXJ,6 = PARMINJ,5 7. At last zero will be the minimum values for the ranges that makes technique six is the

In this section we first present how to calculate the pre-calculated PAR ranges look-up table, which is an essential part in our adaptation algorithm, then we will compare our AHEC technique with the previous works in the literature that depend on only ARQ or only FEC for error correction (Lee et al., 2008; Gao et al., 2009; Tan et al., 2003; Xie and Cui, 2007) in

To calculate the pre-calculated PAR ranges look-up table, energy efficiencies versus PARs for the six techniques are calculated as in section 2.2.3 for ARQ and in section 2.2.4 for the five variable code rate FEC, then the pre-calculated PAR ranges look-up table can be

i\j 1 2 3 4 5 6

2 1.0 0.89 – 1.0 0.84 -0.89 0.62-0.84 0.32-0.62 0. 00-0.32 3 1.0 0.92 -1.0 0.89 -0.92 0.72 -0.89 0.45-0.72 0.00-0.45 4 1.0 0.96 -1.0 0.94 -0.96 0.85 – 0.94 0.68 -.85 0.00 – 0.68 5 1.0 0.98 – 1.0 0.97 – 0.98 0.92– 0.97 0.81–0.92 0.00 -0.81 6 1.0 0.995 – 1.0 0.992 -0.995 0.992 -0.98 0.95 – 0.98 0.00 – 0.95

most energy efficient technique (PARMINJ,6 = 0, for all techniques).

**5.3.1 AHEC technique transmission energy efficiency calculations** 

calculated as in section 3.3, and it can be displayed as in Table 7 below:

PARMAXJ,5 = PARMINJ,4 6. Then decreasing SNR value until the energy efficiency of the fifth technique is less than the energy efficiency of the six technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique 5. This means the PAR of any technique J at this point = PARMINJ,5 i.e. if the PAR of the current technique J is in between PARMINJ,5 and PARMAXJ,5, then technique number 5

#### **5.2 Pre-calculated PAR ranges look-up table calculations**

The pre-calculated PAR ranges look-up table is calculated as follows:


 $\text{PARMAX}$  $\chi\_{1^2} = \text{PARMIN}\_{\text{J.1}}$ 

Then decreasing SNR value until the energy efficiency of the second technique is less than the energy efficiency of the third technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique number two.

This means the PAR of any technique J at this point = PARMINJ,2 i.e. if the PAR of the current technique J is in between PARMINJ,2 and PARMAXJ,2, then technique number 2 is the most energy efficient technique.

As the minimum values in the second ranges equal the maximum values in the third range, then:

#### PARMAXJ,3 = PARMINJ,2

4. Then decreasing SNR value until the energy efficiency of the third technique is less than the energy efficiency of the fourth technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique number 3.

This means the PAR of any technique J at this point = PARMINJ,3 i.e. if the PAR of the current technique J is in between PARMINJ,3 and PARMAXJ,3, then technique number 3 is the most energy efficient technique.

As the minimum values in the third ranges equal the maximum values in the fourth range, then:

#### PARMAXJ,4 = PARMINJ,3

5. Then decreasing SNR value until the energy efficiency of the fourth technique is less than the energy efficiency of the fifth technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique number 4.

This means the PAR of any technique J at this point = PARMINJ,4 i.e. if the PAR of the current technique J is in between PARMINJ,4 and PARMAXJ,4, then technique number 4 is the most energy efficient technique.

As the minimum values in the fourth ranges equal the maximum values in the fifth range, then:

$$\text{PARMAX} \mathbf{X}\_{\text{J.5}} = \text{PARMIN}\_{\text{J.4}}$$

6. Then decreasing SNR value until the energy efficiency of the fifth technique is less than the energy efficiency of the six technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique 5. This means the PAR of any technique J at this point = PARMINJ,5 i.e. if the PAR of the

current technique J is in between PARMINJ,5 and PARMAXJ,5, then technique number 5 is the most energy efficient technique.

As the minimum values in the fifth ranges equal the maximum values in the sixth range, then:

$$\text{PARMAX} \mathbf{X}\_{\text{J},6} = \text{PARMIN}\_{\text{J},5}$$

7. At last zero will be the minimum values for the ranges that makes technique six is the most energy efficient technique (PARMINJ,6 = 0, for all techniques).

#### **5.3 Results and discussion**

64 Energy Efficiency in Communications and Networks

1. Transmission energy efficiencies and PARs using six error correction techniques (One ARQ plus five variable code rates FECs) for variable values of SNR are found as in

2. Starting with the SNR values which gives PAR values equal to 1 for all the techniques; at this SNR ARQ will have the maximum energy efficiency compared to the others, so the PAR for all those technique at this point is the maximum values in the ranges which makes the suitable technique is technique 1 (pure ARQ). This means PARMAXJ,1= 1, i.e. if the current technique is J and the current PAR is in the range that has 1 as the

3. Then decreasing SNR value until the energy efficiency of the first technique is less than the energy efficiency of the second technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique 1 (pure ARQ). This means the PAR of any technique J at this point = PARMINJ,1,i.e. if the PAR of the current technique J is in between PARMINJ,1 and PARMAXJ1, then technique one is the most energy efficient technique. As the minimum values in the first ranges

PARMAXJ,2 = PARMINJ,1

Then decreasing SNR value until the energy efficiency of the second technique is less than the energy efficiency of the third technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is

This means the PAR of any technique J at this point = PARMINJ,2 i.e. if the PAR of the current technique J is in between PARMINJ,2 and PARMAXJ,2, then technique number 2

As the minimum values in the second ranges equal the maximum values in the third

This means the PAR of any technique J at this point = PARMINJ,3 i.e. if the PAR of the current technique J is in between PARMINJ,3 and PARMAXJ,3, then technique number 3

As the minimum values in the third ranges equal the maximum values in the fourth

PARMAXJ,4 = PARMINJ,3 5. Then decreasing SNR value until the energy efficiency of the fourth technique is less than the energy efficiency of the fifth technique; at this SNR the PAR for all technique

PARMAXJ,3 = PARMINJ,2 4. Then decreasing SNR value until the energy efficiency of the third technique is less than the energy efficiency of the fourth technique; at this SNR the PAR for all technique will be the minimum values in the ranges which makes the suitable technique is technique

maximum value, then technique one is the most energy efficient technique.

**5.2 Pre-calculated PAR ranges look-up table calculations** 

equal the maximum values in the second range, then:

section 2.2.3 and section 2.2.4.

technique number two.

range, then:

number 3.

range, then:

is the most energy efficient technique.

is the most energy efficient technique.

The pre-calculated PAR ranges look-up table is calculated as follows:

In this section we first present how to calculate the pre-calculated PAR ranges look-up table, which is an essential part in our adaptation algorithm, then we will compare our AHEC technique with the previous works in the literature that depend on only ARQ or only FEC for error correction (Lee et al., 2008; Gao et al., 2009; Tan et al., 2003; Xie and Cui, 2007) in variable channel conditions and variable distances.

#### **5.3.1 AHEC technique transmission energy efficiency calculations**

To calculate the pre-calculated PAR ranges look-up table, energy efficiencies versus PARs for the six techniques are calculated as in section 2.2.3 for ARQ and in section 2.2.4 for the five variable code rate FEC, then the pre-calculated PAR ranges look-up table can be calculated as in section 3.3, and it can be displayed as in Table 7 below:


Table 7. Pre-Calculated Look-Up PAR Ranges Table

Energy Efficient Communication for Underwater Wireless Sensors Networks 67

AHECT Vs FEC & ARQ Energy Eff. (Variable Channel Conditions)

FEC ARQ AHECT

<sup>0</sup> 0.5 <sup>1</sup> 1.5 <sup>2</sup> 2.5 <sup>3</sup> <sup>0</sup>

Wind Speed (m/s)

Fig. 7. AHEC Technique Vs ARQ & FEC Transmission Energy Efficiency (Variable Channel

increases to 1 m/s, and more than 60 % when wind speed is greater than 1.5 m/s. When compared with transmission using FEC, transmission using AHEC technique achieves around 8 % increase in energy saving when wind speed is below 0.5 m/s and around 6 %

Underwater wireless sensor network (UWSN) is a promising engineering endeavour which will ensure progress in monitoring and exploiting the ocean's vast resources. But until now it faces many challenges, the most important of which is the severe energy constraint of the batteries, which cannot be recharged or replaced in aquatic medium. Complicating the issue is the variability of the channel conditions and the distances

In this chapter, we have mathematically analyzed the transmission energy efficiency for two main error correction techniques, ARQ and FEC, in underwater environment. A simulation is done to validate the mathematical derivation results. Transmission using ARQ is found to be more energy efficient than transmission using FEC below specific distances, and transmission using FEC is found to be better after that. We call this specific distance the cutoff distance. We found that this cut-off distance is not fixed and varies with the variation in

Based on the mathematical analysis, we have proposed an energy efficient Adaptive Hybrid Error Correction (AHEC) technique for transmission. The proposed technique adaptively

0.1

when wind speed is more than 0.5 m/s.

Conditions)

**6. Conclusions** 

between underwater sensors.

channel conditions and packet size.

0.2

0.3

0.4

Energy Eff.

0.5

0.6

0.7

0.8

From the Pre-calculated PAR ranges look-up table above, and from the current PAR, current encoding technique, AHECT energy efficiency can be calculated as in section 3.2

#### **5.3.2 Transmission using AHEC technique versus the transmission using ARQ and FEC transmission energy efficiency**

Figure 6 below gives a comparison between the energy efficiency of transmission using AHEC technique and the transmission using pure ARQ and pure FEC for varying distances.

Fig. 6. AHEC Technique Vs ARQ & FEC Energy Efficiency (Variable Distances Case)

From this figure it is clear that transmission using AHEC technique is more energy efficient than using both ARQ and FEC in variable distances situation.

Compared with the pure ARQ, transmission using AHEC technique achieves 10 % increase in saving energy when the distance is around 1500 m and more than 60 % when the distance increases above 1700 m. When compared with transmission using FEC, it achieves around 10 % increase in energy saving when the distance is below 1500 m, and around 7 % saving when the distance goes above 1500 m.

In Figure 7; variable wind speed is taken as a measure for the variation in channel conditions. From this Figure it is clear that transmission using AHEC technique is more energy efficient than both techniques using ARQ and FEC for variable wind speed (i.e. variable channel conditions). Compared with the pure ARQ, and when the transmission distance is 1500 m, transmission using AHEC technique achieves 6 % increase in energy saving when wind speed is 0.5 m/s, more than 50 % energy saving when wind speed

Fig. 7. AHEC Technique Vs ARQ & FEC Transmission Energy Efficiency (Variable Channel Conditions)

increases to 1 m/s, and more than 60 % when wind speed is greater than 1.5 m/s. When compared with transmission using FEC, transmission using AHEC technique achieves around 8 % increase in energy saving when wind speed is below 0.5 m/s and around 6 % when wind speed is more than 0.5 m/s.

## **6. Conclusions**

66 Energy Efficiency in Communications and Networks

From the Pre-calculated PAR ranges look-up table above, and from the current PAR, current

**5.3.2 Transmission using AHEC technique versus the transmission using ARQ and** 

Figure 6 below gives a comparison between the energy efficiency of transmission using AHEC technique and the transmission using pure ARQ and pure FEC for varying distances.

AHECT Vs FEC & ARQ Energy Efficiencies

<sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>2500</sup> <sup>3000</sup> <sup>0</sup>

Distance (m)

From this figure it is clear that transmission using AHEC technique is more energy efficient

Compared with the pure ARQ, transmission using AHEC technique achieves 10 % increase in saving energy when the distance is around 1500 m and more than 60 % when the distance increases above 1700 m. When compared with transmission using FEC, it achieves around 10 % increase in energy saving when the distance is below 1500 m, and around 7 % saving

In Figure 7; variable wind speed is taken as a measure for the variation in channel conditions. From this Figure it is clear that transmission using AHEC technique is more energy efficient than both techniques using ARQ and FEC for variable wind speed (i.e. variable channel conditions). Compared with the pure ARQ, and when the transmission distance is 1500 m, transmission using AHEC technique achieves 6 % increase in energy saving when wind speed is 0.5 m/s, more than 50 % energy saving when wind speed

Fig. 6. AHEC Technique Vs ARQ & FEC Energy Efficiency (Variable Distances Case)

encoding technique, AHECT energy efficiency can be calculated as in section 3.2

**FEC transmission energy efficiency** 

0.1

when the distance goes above 1500 m.

FEC (Code Rate = 4/5) Pure ARQ AHECT

than using both ARQ and FEC in variable distances situation.

0.2

0.3

0.4

Energy Eff.

0.5

0.6

0.7

0.8

Underwater wireless sensor network (UWSN) is a promising engineering endeavour which will ensure progress in monitoring and exploiting the ocean's vast resources. But until now it faces many challenges, the most important of which is the severe energy constraint of the batteries, which cannot be recharged or replaced in aquatic medium. Complicating the issue is the variability of the channel conditions and the distances between underwater sensors.

In this chapter, we have mathematically analyzed the transmission energy efficiency for two main error correction techniques, ARQ and FEC, in underwater environment. A simulation is done to validate the mathematical derivation results. Transmission using ARQ is found to be more energy efficient than transmission using FEC below specific distances, and transmission using FEC is found to be better after that. We call this specific distance the cutoff distance. We found that this cut-off distance is not fixed and varies with the variation in channel conditions and packet size.

Based on the mathematical analysis, we have proposed an energy efficient Adaptive Hybrid Error Correction (AHEC) technique for transmission. The proposed technique adaptively

Energy Efficient Communication for Underwater Wireless Sensors Networks 69

Lee, W. Kim, J. Lee, J. Jang, Y and Dho, K. (2008). An improved ARQ scheme in underwater

LinkQuest Inc. Underwater Acoustic Modem Models. June 2010 (Available:

Liu, B. Chen, H. Lei, X. Ren, F. and Sezaki , K. (2010). Internode distance-based redundancy

Lin S., Costello D., Miller M., (1984) Automatic-Repeat-Request Error-Control Schemes.

Pompili, D. (2007). *Efficient communication protocols for underwater acoustic sensor networks*,"

Preisig, J. (2007). Acoustic Propagation considerations for underwater Acoustic

S. Joshy and A. Babu, "Capacity of Underwater Wireless Communication Channel with

Sankarasubramaniam, Y. Akyidiliz, I. and Mclaughlin, S. (2003). Energy efficiency based

Tan H., Seah W. and Doyle L. (2007). A multi-hop ARQ protocol for underwater acoustic networks, Proceedings of OCEANS 2007 , Europe, 18-21 June 2007 pp. 1 - 6. Tian, Z. Yan, D. and Liang, Q. (2008). Energy efficiency analysis of error control schemes in

Valera, A. Lee, P. Tan, H. (2009). Implementation and Evaluation of Multihop ARQ for

Webb, S. (1992). The equilibrium oceanic microseism spectrum, J. Acoust. Soc., Vol. 92, pp.

Wenz and Gordon, M. (1936). Acoustic Ambient Noise in the Ocean: Spectra and Sources,

Xie P. (2007). Underwater acoustic sensor networks: Medium access control, routing and

Xie, P. and Cui, J. (2007). An FEC- based reliable data transport protocol for Underwater

sensor networks, Proc. of 16th International Conference on Computer

and Mobile Computing Conference, IWCMC '08, pp. (401-405), 2008.

Urick R. J. (1983). Principles of Underwater Sound, McGraw-Hill, New York.

The Journal of the Acoustical Society of America, Vol. 34, 1936.

IEEE OCEANS Conference, Bremen, Germany, 2009.

Communications and Networks, pp.( 747-753), 2007.

(held in conjunction with IEEE ICC'03), Anchorage, Alaska, USA, 2003. Stojanovic, M. and Preisig. J. (2009). Underwater Acoustic Communication Channels:

2008.

http://www.link-quest.com)

Rev., Vol. 11, pp. (2-10), 2007.

IEEE Communications Magazine. pp (5-17), 1984.

Networks & Communications (IJCNC), vol. 2, 2010.

Networking, 2010.

Technology, 2007.

84-89, 2009.

(2141-2158), 1992.

reliable transfer, 2007.

acoustic sensor networks, Proceeding of MTS/IEEE Oceans, Kobe, Japan, pp. (1-5),

reliable transport in underwater sensor networks, EURASIP J. Wireless Comm. and

PhD, School of Electrical and Computer Engineering, Georgia Institute of

communications network Development , SIGMOBILE Mob. Computer Commun.

Different Acoustic Propoagation Loss Models," International Journal of Computer

packet size optimization in wireless sensor networks , Proceeding of the 1st IEEE International Workshop on Sensor Network Protocols and Applications SNPA'03

Propagation Models and Statistical Characterization, IEEE Communications, pp.

wireless sensor networks, Proceeding of International Wireless Communications

Reliable Communications in Underwater Acoustic Networks, Proceedings of the

changes the error correction technique to the technique with the highest transmission energy efficiency compared to the others. An adaptation algorithm which based on the current packet acceptance rate (PAR), current encoding technique, and a pre-calculated PAR ranges look-up table has been proposed. From the output of the adaptation algorithm, a periodic 3 bit feedback is sent to the sender indicating which error correction technique is most suitable given the current distance and channel conditions. The proposed technique has been compared with techniques that use only ARQ or FEC. The results show that our proposed technique is more energy efficient than either of them.

#### **7. References**


68 Energy Efficiency in Communications and Networks

changes the error correction technique to the technique with the highest transmission energy efficiency compared to the others. An adaptation algorithm which based on the current packet acceptance rate (PAR), current encoding technique, and a pre-calculated PAR ranges look-up table has been proposed. From the output of the adaptation algorithm, a periodic 3 bit feedback is sent to the sender indicating which error correction technique is most suitable given the current distance and channel conditions. The proposed technique has been compared with techniques that use only ARQ or FEC. The results show that our

Akyildiz, I. Pompili, D. and Melodia, T. (2004). Challenges for Efficient Communication in Underwater Acoustic Sensor Networks, *ACM Sigbed Review*, Vol. 1, 2004. Akyildiz, I. Pompili, D. and Melodia, T. (2005). Underwater acoustic sensor networks: research challenge, Ad Hoc Networks (Elsevier), Vol. 3, pp. (257-279), 2005. Begin, G. Haccoun, D. and Paquin, C. (1990). Further Results on High - Rate Puncturing G

Bin, L. Garcin, F. Ren, F. and Lin, C. (2008). A study of forward error correction schemes for

Gao, M. Soh, W. and Tao, M. (2009). A transmission scheme for continuous ARQ protocols

Guo, Z. Xie, P. Cui, J. and Wang, B. (2006). On applying network coding to underwater sensor networks, Proceedings of ACM WUWNet'06, Los Angeles, CA, 2006. Harris, A. and Zorzi, M. (2007). Modelling the underwater acoustic channel in ns2,

Joshy, S. and Babu, A. (2010). Capacity of Underwater Wireless Communication Channel

Kunal K., Tripathi R. and Singh V. (2010). An HARQ based Optimized Error Correction Technique, International Journal of Computer Applications, Vol. 9, 2010. Labrador, Y. Karimi, M. Pan, D. And Miller, J. (2009). Modulation and error correction in the

Lee J. Kim J. Lee J. Jang Y. and Dho K. (2008). An improved ARQ scheme in underwater acoustic sensor networks, Proc. MTS/IEEE Oceans, Kobe, Japan, 2008, pp. 1-5.

Communications and Networks (SECON '08), pp (197–205), 2008. Brekhovskikh, L. and Lysanov, Y. (2003). Fundamental of ocean acoustics: Springer, 1982. Colin Y., Chan, and Motani M. (2007). An integrated energy efficient data retrieval protocol

Convolutional codes for viterbi and sequential decoding, IEEE transaction on

reliable transport in underwater sensor networks, Proceedings of the 5th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc

for underwater delay tolerant networks, Proceeding of IEEE ocean'07, Aberdeen,

over underwater acoustic channels, Proceeding of the 2009 IEEE international

Proceedings of the 2nd international conference on Performance evaluation

with Different Acoustic Propoagation Loss Models, *International Journal of Computer* 

underwater acoustic communication channel, International Journal of Computer

proposed technique is more energy efficient than either of them.

Communications, Vol. 38, pp. (1922-1928), 1990.

conference on Communications (ICC'09), 2009.

*Networks & Communications (IJCNC)*, Vol. 2, 2010.

Science and Network Security, Vol. 9, pp. (123-130), 2009.

methodologies and tools, 2007.

**7. References** 

Scotland, 2007.


**0**

**4**

*USA*

**Energy Efficiency of Connected Mobile Platforms**

In the last decade there has been an explosive growth in popularity of mobile computing platforms which include laptops, notebooks, tablets, cell phones, etc. However, the usability of these devices from end users point of view is directly associated with their battery life and the fact that they are powered by non-continuous energy sources imposes a serious limitations to these devices. As a result, a typical architecture design of such mobile computing platforms will include low power platform states for individual components (e.g. CPU, memory controller, hard-disk, etc.) or for the whole platform. These low power states can be *sleep* states where the components or the whole platform is in a non-operational mode (e.g. operating systems defined sleep states: standby, hibernate, etc.) or *scaled* states where they are operating at lower than peak performance (e.g. CPU dynamic voltage and frequency scaling (DVFS)). For example the Advanced Configuration and Power Interface (ACPI) specifies power management concepts and interfaces. ACPI integrates the operating system, device drivers, system hardware components, and applications for power management and defines several power states for each component ranging from fully powered on to fully powered-off

Individual power management techniques differ in (1) how deep the low power state is, where usually a deeper sleep state has a higher exit latency and hence a negative performance impact, (2) the algorithms used to manage entering and exiting these low power states and (3) optimizations to extend these sleep states as long as possible so additional system components having longer exit latency could also be put into lower power state. Standard policies are usually based on *timeout* approaches, where the platform or individual components are transitioned in low-power modes if they have been observed "idle" for the duration of a timeout. Usually this timeout value is dynamic and depends on the current operating state

Recent measurements and field assessments have shown that: in offices, desktop systems at a minimum remain powered up all day long whether being actively used or not, and further, two-thirds of such systems are fully on after work hours, with only 4% operating in sleep modes. It appears plausible that network connectivity drives much of this and these systems remain available to facilitate sporadic, occasional activity, such as user remote access, administrator access for maintenance, etc. (backups, patch management). On the other hand, in the residential sector, the average PC is on 34% of the time and spends only 4% of the time

with each successive state consuming the same or less amount of power.

and the previous history of idle and active platform states.

**1. Introduction**

**in Presence of Background Traffic**

*Circuits and Systems Research Lab, Intel Labs, Intel Corporation*

Sameh Gobriel, Christian Maciocco and Tsung-Yuan Charlie Tai

Yang H. and Liu B. (2009). Optimization of Energy Efficient Transmission in Underwater Sensor Networks, Proceedings of CoRR, 2009.

## **Energy Efficiency of Connected Mobile Platforms in Presence of Background Traffic**

Sameh Gobriel, Christian Maciocco and Tsung-Yuan Charlie Tai *Circuits and Systems Research Lab, Intel Labs, Intel Corporation USA*

#### **1. Introduction**

70 Energy Efficiency in Communications and Networks

Yang H. and Liu B. (2009). Optimization of Energy Efficient Transmission in Underwater

Sensor Networks, Proceedings of CoRR, 2009.

In the last decade there has been an explosive growth in popularity of mobile computing platforms which include laptops, notebooks, tablets, cell phones, etc. However, the usability of these devices from end users point of view is directly associated with their battery life and the fact that they are powered by non-continuous energy sources imposes a serious limitations to these devices. As a result, a typical architecture design of such mobile computing platforms will include low power platform states for individual components (e.g. CPU, memory controller, hard-disk, etc.) or for the whole platform. These low power states can be *sleep* states where the components or the whole platform is in a non-operational mode (e.g. operating systems defined sleep states: standby, hibernate, etc.) or *scaled* states where they are operating at lower than peak performance (e.g. CPU dynamic voltage and frequency scaling (DVFS)). For example the Advanced Configuration and Power Interface (ACPI) specifies power management concepts and interfaces. ACPI integrates the operating system, device drivers, system hardware components, and applications for power management and defines several power states for each component ranging from fully powered on to fully powered-off with each successive state consuming the same or less amount of power.

Individual power management techniques differ in (1) how deep the low power state is, where usually a deeper sleep state has a higher exit latency and hence a negative performance impact, (2) the algorithms used to manage entering and exiting these low power states and (3) optimizations to extend these sleep states as long as possible so additional system components having longer exit latency could also be put into lower power state. Standard policies are usually based on *timeout* approaches, where the platform or individual components are transitioned in low-power modes if they have been observed "idle" for the duration of a timeout. Usually this timeout value is dynamic and depends on the current operating state and the previous history of idle and active platform states.

Recent measurements and field assessments have shown that: in offices, desktop systems at a minimum remain powered up all day long whether being actively used or not, and further, two-thirds of such systems are fully on after work hours, with only 4% operating in sleep modes. It appears plausible that network connectivity drives much of this and these systems remain available to facilitate sporadic, occasional activity, such as user remote access, administrator access for maintenance, etc. (backups, patch management). On the other hand, in the residential sector, the average PC is on 34% of the time and spends only 4% of the time

Recent work has shown that Smart timing approaches for Operating systems (OS) can be used to increase the quiet times for the OS by skipping timer tick interrupts when the platform is idle or by adaptively changing the rate of timer interrupts (e.g., Olsen & Narayanaswami (2006)) which form the basis for *"tickless OS"* where the OS is moving away from scheduling periodic clock interrupts every few milliseconds to a more event-driven approach Gleixner & Molnar (2006); Yodaiken & Barabanov (1997) where the OS is waken to process an event being

<sup>73</sup> Energy Efficiency of

Figure 1 represents the energy gain of platform power management and typically the achieved gain follows a non-decreasing concave reward function Olsen & Narayanaswami (2006); Siddha et al. (2007). The x-axis represents the break event time, i.e., guaranteed platform quietness period, and the y-axis represents the normalized increase in the total battery life when the system enters a low power state normalized to the case when it remains in the same fully active state (S0). The desired operating range is highlighted and extending the sleeping time beyond the highlighted range has a diminishing return in the energy gain as it is not possible to extend the tick rate for the OS beyond the highlighted range Olsen & Narayanaswami (2006) and the system will be wakened for some scheduled event at about

When a system is connected to a network, even when the user is not engaged in any active communication the system engages in what we term "background/noise" traffic. These packets can be network maintenance or management packets (e.g., ARP, DHCP, etc.), network services handshake (e.g., directory services, NetBIOS) or application heartbeats (e.g., Instant Messaging, Windows Widgets, etc.). Figures 2 and 3 show the average and variance in the number of network packets processed per second and the protocol mix of two examples of this background traffic. Figure 2 is a wired network trace for a home environment and Figure 3 is

It should be clear that these network traces are shown here as examples of real-life networks and cannot be generalized as the default background traffic in these networks. As previously mentioned the traffic characteristics will be remarkably different from location to location based on the network infrastructure architecture, platforms and applications used, number of network users, time of data collection etc. In fact chattier and less chatty background

posted by the applications or by the hardware on demand.

Connected Mobile Platforms in Presence of Background Traffic

Fig. 1. Typical Platform PM Energy Gain

**3.2 Background traffic effect on platform PM**

a wireless trace for a platform connected to an enterprise network.

this range.

in an some form of sleep state and more than half the time a PC is on no one is actively using the machine.

In this chapter we will argue that a major hurdle for end systems desiring to enter a sleep state arises from the network and the desire of end users to stay *"always on and always connected"* to the network. We first focus on powered-on platforms but idle (i.e. not running an active heavy workload) and quantify the effect of background network traffic on the platform energy-efficiency. then, we propose a low-overhead mitigation techniques that can detect and classify network traffic with no dependencies on network protocols nor the platform and present simulation analysis of these techniques and algorithms based on live network traces. Finally we describe our firmware based implementation in our prototype Wireless Network Interface Cards (WNIC) and show our practical results when applied in live networks.

#### **2. Related work**

Several mechanisms have been proposed to reduce the energy consumption of networked platforms. Prior work can largely be grouped in three categories: reducing the active power consumption of systems when they are awake Agarwal et al. (2008); Flinn & Satyanarayanan (2004); Li et al. (2007), reducing the power consumption of the network infrastructure, routers and switches Gunaratne et al. (2005); Gupta & Singh (2003); Nedevschi et al. (2008) and opportunistically putting the devices to sleep Agarawal et al. (2009); Shih et al. (2002); Sorber et al. (2005). *Long Idle* falls in the third category, where it advocates for localized energy-efficient optimization within the platform to extend the sleeping state of the platform rather than using a network-wide implementation of a proxy (wakeup) service. However, in contrast with previous work *Long idle* did not assume a special magic packet for wake-on WLan (WoWLAN) Shih et al. (2002) nor it uses multiple network interface cards Sorber et al. (2005), and it did not assume any application level optimization to offload background traffic processing Agarawal et al. (2009). *Long idle* requires only a single network interface card and detects background traffic without the need to decrypt the data payload by gathering traffic statistics (size, direction, interarrival time, etc.) on the ongoing communication. In this chapter, we show that such non intrusive technique achieves a good detection accuracy and extends the platform sleeping state with no need for a more sophisticated technique or a special hardware.

#### **3. Problem formulation**

In this section, we analyze and quantify the negative impact of continuously interrupting the platform to process the background and management network traffic and highlight the effect on the platform energy-efficiency.

#### **3.1 Platform PM energy gain**

Current platform power management guides the platform to enter a lower power state (sleeping state Sx) if it detects a period of idleness; *τidle*. The more the system stays in sleep state un-interrupted the better the energy gain is, because more and more peripherals can enter a low power state and/or because deeper sleep states with longer exit latency can be used.

2 Will-be-set-by-IN-TECH

in an some form of sleep state and more than half the time a PC is on no one is actively using

In this chapter we will argue that a major hurdle for end systems desiring to enter a sleep state arises from the network and the desire of end users to stay *"always on and always connected"* to the network. We first focus on powered-on platforms but idle (i.e. not running an active heavy workload) and quantify the effect of background network traffic on the platform energy-efficiency. then, we propose a low-overhead mitigation techniques that can detect and classify network traffic with no dependencies on network protocols nor the platform and present simulation analysis of these techniques and algorithms based on live network traces. Finally we describe our firmware based implementation in our prototype Wireless Network Interface Cards (WNIC) and show our practical results when applied in live networks.

Several mechanisms have been proposed to reduce the energy consumption of networked platforms. Prior work can largely be grouped in three categories: reducing the active power consumption of systems when they are awake Agarwal et al. (2008); Flinn & Satyanarayanan (2004); Li et al. (2007), reducing the power consumption of the network infrastructure, routers and switches Gunaratne et al. (2005); Gupta & Singh (2003); Nedevschi et al. (2008) and opportunistically putting the devices to sleep Agarawal et al. (2009); Shih et al. (2002); Sorber et al. (2005). *Long Idle* falls in the third category, where it advocates for localized energy-efficient optimization within the platform to extend the sleeping state of the platform rather than using a network-wide implementation of a proxy (wakeup) service. However, in contrast with previous work *Long idle* did not assume a special magic packet for wake-on WLan (WoWLAN) Shih et al. (2002) nor it uses multiple network interface cards Sorber et al. (2005), and it did not assume any application level optimization to offload background traffic processing Agarawal et al. (2009). *Long idle* requires only a single network interface card and detects background traffic without the need to decrypt the data payload by gathering traffic statistics (size, direction, interarrival time, etc.) on the ongoing communication. In this chapter, we show that such non intrusive technique achieves a good detection accuracy and extends the platform sleeping state with no need for a more sophisticated technique or a

In this section, we analyze and quantify the negative impact of continuously interrupting the platform to process the background and management network traffic and highlight the effect

Current platform power management guides the platform to enter a lower power state (sleeping state Sx) if it detects a period of idleness; *τidle*. The more the system stays in sleep state un-interrupted the better the energy gain is, because more and more peripherals can enter a low power state and/or because deeper sleep states with longer exit latency can be

the machine.

**2. Related work**

special hardware.

used.

**3. Problem formulation**

on the platform energy-efficiency.

**3.1 Platform PM energy gain**

Recent work has shown that Smart timing approaches for Operating systems (OS) can be used to increase the quiet times for the OS by skipping timer tick interrupts when the platform is idle or by adaptively changing the rate of timer interrupts (e.g., Olsen & Narayanaswami (2006)) which form the basis for *"tickless OS"* where the OS is moving away from scheduling periodic clock interrupts every few milliseconds to a more event-driven approach Gleixner & Molnar (2006); Yodaiken & Barabanov (1997) where the OS is waken to process an event being posted by the applications or by the hardware on demand.

Fig. 1. Typical Platform PM Energy Gain

Figure 1 represents the energy gain of platform power management and typically the achieved gain follows a non-decreasing concave reward function Olsen & Narayanaswami (2006); Siddha et al. (2007). The x-axis represents the break event time, i.e., guaranteed platform quietness period, and the y-axis represents the normalized increase in the total battery life when the system enters a low power state normalized to the case when it remains in the same fully active state (S0). The desired operating range is highlighted and extending the sleeping time beyond the highlighted range has a diminishing return in the energy gain as it is not possible to extend the tick rate for the OS beyond the highlighted range Olsen & Narayanaswami (2006) and the system will be wakened for some scheduled event at about this range.

#### **3.2 Background traffic effect on platform PM**

When a system is connected to a network, even when the user is not engaged in any active communication the system engages in what we term "background/noise" traffic. These packets can be network maintenance or management packets (e.g., ARP, DHCP, etc.), network services handshake (e.g., directory services, NetBIOS) or application heartbeats (e.g., Instant Messaging, Windows Widgets, etc.). Figures 2 and 3 show the average and variance in the number of network packets processed per second and the protocol mix of two examples of this background traffic. Figure 2 is a wired network trace for a home environment and Figure 3 is a wireless trace for a platform connected to an enterprise network.

It should be clear that these network traces are shown here as examples of real-life networks and cannot be generalized as the default background traffic in these networks. As previously mentioned the traffic characteristics will be remarkably different from location to location based on the network infrastructure architecture, platforms and applications used, number of network users, time of data collection etc. In fact chattier and less chatty background

(a) Traffic Rate

<sup>75</sup> Energy Efficiency of

(b) Protocol Mix

Moreover, encountering an additional performance guard timeout before re-entering the low power state will degrade the net energy gain of platform power management even further

Figure 5(a) plots the Cumulative distribution function of uninterrupted packet free times in different environments. As shown in figure in an enterprise environment (traffic trace example shown in Figure 3) almost 40% of the time the platform will be hit by a packet within 100 msec and in a wired home environment (traffic trace example shown in Figure 2) the probability is about 60%. The platform power management energy gain degrades accordingly with background network traffic. If we assume that the platform activities are synchronized and an uninterrupted sleep time of 100 msec is guaranteed each time the platform enters

Fig. 3. Enterprise Wireless Network Background Traffic

Connected Mobile Platforms in Presence of Background Traffic

Fig. 4. Traffic Effect on PM Energy Gain

(for illustration *τidle* = 50 msec in figure).

Fig. 2. Home Wired Network Background Traffic

and management network traffic has been reported in different literatures Gunaratne et al. (2005). One of the significance of our work is that it is not designed to target specific network conditions and no specific background/management traffic network optimization (e.g., ARP-Proxy Nedevschi et al. (2009), Network Proxy Offload *TC38-TG4 Proxying support for sleep modes Specifications* (2009), etc.) is assumed.

For a networked platform keeping the sleep state of the system uninterrupted is a challenge. Network activity will interrupt the sleeping state because the host is waken-up to process these packets. Moreover, each time the sleeping state is interrupted the platform will have to wait for the timeout (i.e., *τidle*) before re-entering the low power state.

It should be noted however, that *τidle* can't be a small value because this will have a negative impact on active workload due to the overhead (entry and exit latency and power overhead) associated with transitioning to and from the low power managed state. For example if a Wireless NIC is used and according to the 802.11 Power Saving Mode (PSM) *IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications* (1999) each time the NIC enters the sleep mode, it has to wait for the next beacon advertised by the base station before it can receive any of its buffered packets in the base station. Typically the beacon interval is set at 100 msec which means that the exit latency from the sleep state of the wireless NIC in PSM is 100 msec. Hence, if the sleep state is entered blindly each time there is packet-free time the active communication performance (e.g., throughput, latency, etc.) will significantly degrade.

The net effect of background traffic can be viewed as depicted in Figure 4 (dotted red line). If the interpacket arrival time is *X* msec then longer platform sleep times are not available and thus the energy gain of platform power management beyond some value is not achievable. 4 Will-be-set-by-IN-TECH

(a) Traffic Rate

(b) Protocol Mix

and management network traffic has been reported in different literatures Gunaratne et al. (2005). One of the significance of our work is that it is not designed to target specific network conditions and no specific background/management traffic network optimization (e.g., ARP-Proxy Nedevschi et al. (2009), Network Proxy Offload *TC38-TG4 Proxying support*

For a networked platform keeping the sleep state of the system uninterrupted is a challenge. Network activity will interrupt the sleeping state because the host is waken-up to process these packets. Moreover, each time the sleeping state is interrupted the platform will have to

It should be noted however, that *τidle* can't be a small value because this will have a negative impact on active workload due to the overhead (entry and exit latency and power overhead) associated with transitioning to and from the low power managed state. For example if a Wireless NIC is used and according to the 802.11 Power Saving Mode (PSM) *IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications* (1999) each time the NIC enters the sleep mode, it has to wait for the next beacon advertised by the base station before it can receive any of its buffered packets in the base station. Typically the beacon interval is set at 100 msec which means that the exit latency from the sleep state of the wireless NIC in PSM is 100 msec. Hence, if the sleep state is entered blindly each time there is packet-free time the active communication performance (e.g., throughput, latency, etc.) will

The net effect of background traffic can be viewed as depicted in Figure 4 (dotted red line). If the interpacket arrival time is *X* msec then longer platform sleep times are not available and thus the energy gain of platform power management beyond some value is not achievable.

Fig. 2. Home Wired Network Background Traffic

*for sleep modes Specifications* (2009), etc.) is assumed.

significantly degrade.

wait for the timeout (i.e., *τidle*) before re-entering the low power state.

Fig. 3. Enterprise Wireless Network Background Traffic

Fig. 4. Traffic Effect on PM Energy Gain

Moreover, encountering an additional performance guard timeout before re-entering the low power state will degrade the net energy gain of platform power management even further (for illustration *τidle* = 50 msec in figure).

Figure 5(a) plots the Cumulative distribution function of uninterrupted packet free times in different environments. As shown in figure in an enterprise environment (traffic trace example shown in Figure 3) almost 40% of the time the platform will be hit by a packet within 100 msec and in a wired home environment (traffic trace example shown in Figure 2) the probability is about 60%. The platform power management energy gain degrades accordingly with background network traffic. If we assume that the platform activities are synchronized and an uninterrupted sleep time of 100 msec is guaranteed each time the platform enters

(a) Cost for L2 Packets

<sup>77</sup> Energy Efficiency of

Connected Mobile Platforms in Presence of Background Traffic

(b) Cost for L2+ Packets

Figure 6(b) plots the processing cost of L2+ packets (all the packets that are traced in Figures 2 and 3). As shown in figure each of the L2+ packets cause an interrupt reflecting a memory DMA for the packet data followed by two DPCs one is caused by the NIC driver running and the other by the TCP/IP OS stack running. The running time of each depends on whether the packet is dropped or is processed but similarly what we want to highlight is that the host is

We define an architecture and a low-overhead technique, which we call *Long Idle* to enable the NIC to classify incoming packets into active versus background traffic and intelligently hold the background traffic on the NIC device to guarantee idleness so that the system can

Fig. 6. Processing Cost for Background Packets

**4. Long Idle technology**

interrupted at least once to process each of these packets.

(b) Energy Gain @ 100 msec

#### Fig. 5. Platform PM Gain With Network Traffic

the sleep state then battery life is extended by almost 88%. On the other hand when the platform is associated with the network and is processing the background network traffic the gain diminishes (even with *τidle* = 0 msec) to only 56% for the example wireless enterprise network and to about 27% for the wired home environment as shown in Figure 5(b)

#### **3.3 Background traffic processing cost**

We use Windows Performance Tools (WPT) Kit *Windows Performance Analysis Tools* (2008) designed for measuring and analyzing system and application performance on Windows to quantify the processing cost of each packet of background traffic at the different layers of the networking stack and to highlight that each packet will generate at least one interrupt which will wake the platform up to exit the low power platform state.

Figure 6(a) plots the processing cost of Layer 2 (L2) packets. It shows the interrupts and the Deferred Procedure Calls (DPC) generated by the NIC when L2 packets are processed. As shown in the figure when beacons are received the NIC generates two interrupts reflecting information (TIM info, SNR data, etc.) being DMA-ed into the host memory these two interrupts are spaced within approximately 0.5 msec. Each interrupt is followed by a DPC reflecting that the NIC driver is running. Investigating the shape and reasons of these interrupts is beyond the scope of this chapter but we highlight that for each of these interrupts or the DPC the CPU is running (C0 at 800 MHz) indicating that host is wakened up to process each of them.

6 Will-be-set-by-IN-TECH

(a) CDF of Packet Free Time

(b) Energy Gain @ 100 msec

the sleep state then battery life is extended by almost 88%. On the other hand when the platform is associated with the network and is processing the background network traffic the gain diminishes (even with *τidle* = 0 msec) to only 56% for the example wireless enterprise

We use Windows Performance Tools (WPT) Kit *Windows Performance Analysis Tools* (2008) designed for measuring and analyzing system and application performance on Windows to quantify the processing cost of each packet of background traffic at the different layers of the networking stack and to highlight that each packet will generate at least one interrupt which

Figure 6(a) plots the processing cost of Layer 2 (L2) packets. It shows the interrupts and the Deferred Procedure Calls (DPC) generated by the NIC when L2 packets are processed. As shown in the figure when beacons are received the NIC generates two interrupts reflecting information (TIM info, SNR data, etc.) being DMA-ed into the host memory these two interrupts are spaced within approximately 0.5 msec. Each interrupt is followed by a DPC reflecting that the NIC driver is running. Investigating the shape and reasons of these interrupts is beyond the scope of this chapter but we highlight that for each of these interrupts or the DPC the CPU is running (C0 at 800 MHz) indicating that host is wakened up to process

network and to about 27% for the wired home environment as shown in Figure 5(b)

will wake the platform up to exit the low power platform state.

Fig. 5. Platform PM Gain With Network Traffic

**3.3 Background traffic processing cost**

each of them.

Fig. 6. Processing Cost for Background Packets

Figure 6(b) plots the processing cost of L2+ packets (all the packets that are traced in Figures 2 and 3). As shown in figure each of the L2+ packets cause an interrupt reflecting a memory DMA for the packet data followed by two DPCs one is caused by the NIC driver running and the other by the TCP/IP OS stack running. The running time of each depends on whether the packet is dropped or is processed but similarly what we want to highlight is that the host is interrupted at least once to process each of these packets.

#### **4. Long Idle technology**

We define an architecture and a low-overhead technique, which we call *Long Idle* to enable the NIC to classify incoming packets into active versus background traffic and intelligently hold the background traffic on the NIC device to guarantee idleness so that the system can

1. Idle traffic is mostly small-sized packets (typically a heartbeat or a management packet is

<sup>79</sup> Energy Efficiency of

2. Because most of the management and background traffic is transmitted by the network,

3. Active small-sized packets (e.g., those belong to a VoIP or a gaming session) are mostly

4. A large percentage of the Idle traffic is broadcast packets, while active communication is

Our algorithm for enabling *Long Idle* is composed of two state machines. One is continuously sampling the ongoing traffic and classifies whether it is active versus background at the end of the sampling window and is independent from the NIC used (wired or wireless), the other is defining how to handle the packet based on the classification result and clearly this depends on the type of the NIC used. In this chapter, we use the Wireless NIC as an example and show how the packet handling can be coupled with 802.11 PSM u-apsd *IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications* (1999) mode.

As shown in Figure 8, the algorithm for traffic sampling and *Long Idle* divides the time into slices of sampling windows. During this sampling window statistics on the size,

only few tens of bytes) and MTU size packets are rarely seen in idle traffic.

idle traffic is mostly on the receive path.

Fig. 8. Long Idle Sampling State Machine

mostly sending and receiving unicast packets.

Connected Mobile Platforms in Presence of Background Traffic

two-way communication.

enter low power states with no impact on performance. It should be noted, however, that although we present *Long Idle* technology as something that we implemented in the NIC there is nothing that prevents the network designer from applying that same technique into for example, network switches and/or network wireless access points.

#### **4.1 Long Idle overview**

By definition background packets are not for active communication, hence, typically these packets are not time critical and can be delayed for extended period of time without any impact on the user experience. In fact most of these background packets are discarded by the host after processing them Nedevschi et al. (2009; 2008).

Unfortunately, at the NIC level it is not possible (without performing deep inspection of the packet) to know whether the packet will be of any use to the host or not. Since for all practical reasons the time and processing overhead and energy of deep packet inspection is intolerable at that NIC level if at all possible due security and encryption (e.g., VPN, etc.), the hypothesis of *Long Idle* is: if the NIC can classify the incoming packets into active versus background traffic without having to look inside the packet then the NIC can intentionally buffer the background packets and guarantee uninterrupted quietness to the platform for extended amount of time, e.g., few hundreds msec, and then send these packets as one burst to the host for processing. Figure 7 depicts the *Long Idle* concept. Moreover, since the created quietness period is guaranteed not to be interrupted except for active traffic then the platform can *instantly* enter a sleep state for the whole duration of the buffering period (without waiting for *τidle*) to save its energy even further.

Fig. 7. Long Idle Overview

#### **4.2 Long Idle algorithm**

Once a packet is received at the NIC and is classified as belonging to background traffic the packet is not instantly pushed to the host, but instead buffered inside the NIC's internal memory until one of few events occurs: Either, a *Long Idle* timeout event has occurred (maximum time watermark (e.g., 300 msec) for holding a background packet is reached) or a packet belonging to an active communication is received in the Rx FIFO or the Rx FIFO has reached some occupancy threshold (maximum bytes to hold for background traffic).

*Long Idle* relies on the NIC's ability to differentiate the active traffic from the idle background one without a need for packet deep inspection. Based on our analysis the main differences in the traffic patterns can be summarized as follows:

8 Will-be-set-by-IN-TECH

enter low power states with no impact on performance. It should be noted, however, that although we present *Long Idle* technology as something that we implemented in the NIC there is nothing that prevents the network designer from applying that same technique into for

By definition background packets are not for active communication, hence, typically these packets are not time critical and can be delayed for extended period of time without any impact on the user experience. In fact most of these background packets are discarded by the

Unfortunately, at the NIC level it is not possible (without performing deep inspection of the packet) to know whether the packet will be of any use to the host or not. Since for all practical reasons the time and processing overhead and energy of deep packet inspection is intolerable at that NIC level if at all possible due security and encryption (e.g., VPN, etc.), the hypothesis of *Long Idle* is: if the NIC can classify the incoming packets into active versus background traffic without having to look inside the packet then the NIC can intentionally buffer the background packets and guarantee uninterrupted quietness to the platform for extended amount of time, e.g., few hundreds msec, and then send these packets as one burst to the host for processing. Figure 7 depicts the *Long Idle* concept. Moreover, since the created quietness period is guaranteed not to be interrupted except for active traffic then the platform can *instantly* enter a sleep state for the whole duration of the buffering period (without waiting

Once a packet is received at the NIC and is classified as belonging to background traffic the packet is not instantly pushed to the host, but instead buffered inside the NIC's internal memory until one of few events occurs: Either, a *Long Idle* timeout event has occurred (maximum time watermark (e.g., 300 msec) for holding a background packet is reached) or a packet belonging to an active communication is received in the Rx FIFO or the Rx FIFO has

*Long Idle* relies on the NIC's ability to differentiate the active traffic from the idle background one without a need for packet deep inspection. Based on our analysis the main differences in

reached some occupancy threshold (maximum bytes to hold for background traffic).

example, network switches and/or network wireless access points.

host after processing them Nedevschi et al. (2009; 2008).

for *τidle*) to save its energy even further.

Fig. 7. Long Idle Overview

the traffic patterns can be summarized as follows:

**4.2 Long Idle algorithm**

**4.1 Long Idle overview**


Fig. 8. Long Idle Sampling State Machine

Our algorithm for enabling *Long Idle* is composed of two state machines. One is continuously sampling the ongoing traffic and classifies whether it is active versus background at the end of the sampling window and is independent from the NIC used (wired or wireless), the other is defining how to handle the packet based on the classification result and clearly this depends on the type of the NIC used. In this chapter, we use the Wireless NIC as an example and show how the packet handling can be coupled with 802.11 PSM u-apsd *IEEE 802.11, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications* (1999) mode.

As shown in Figure 8, the algorithm for traffic sampling and *Long Idle* divides the time into slices of sampling windows. During this sampling window statistics on the size,

**5.1 Traffic categorization accuracy**

algorithms to quantify the accuracy of these algorithms.

Connected Mobile Platforms in Presence of Background Traffic

Fig. 10. Long Idle Classification Algorithm Accuracy

in the NIC and processing is delayed).

*Long Idle* relies on the NIC's ability to categorize the network traffic based on the conditions and algorithms outlined in Section 4. We use real-network traffic traces (as mentioned in Section 3) that represent hours of idle and active platform network activities as an input to our *Long idle* simulator running the classification and traffic-type-specific data handling

<sup>81</sup> Energy Efficiency of

(a) Wired Environment

(b) Wireless Environment

Figure 10 analyzes the accuracy of the traffic classification algorithm in two environments wired home environment and wireless enterprise environment. The trend of this result is the same for all other tested networks. The false positive ratio represents the percentage of time idle traffic has been classified as active (i.e., the platform is waken up unnecessarily to process these idle packets) to the total time until it is correctly classified as idle. The false negative ratio represents the percentage of time active traffic is classified as idle (i.e., packets are held

Figure 10 illustrates the effectiveness of our traffic categorization algorithm. As shown in the figure the average accuracy of the algorithm at a sampling window size of 1 sec in an enterprise environment is about 92% and in a home environment is close to 96%. The false negative percentage is very small which indicates that active traffic is never classified as idle, as a result active traffic is not buffered in the NIC and hence the active traffic performance and the user experience are unaffected. As shown in the figure most of the classification errors are

Fig. 9. Long Idle WNIC Implementation

destination, interarrival times, and number of transmitted and received packets are collected and based on the conditions outlined above the ongoing communication is classified as active versus background one. Figure 9 depicts the wireless NIC sleep time update based on the classification of the ongoing traffic. If *Long Idle* is detected the Wireless NIC sleeps longer and the received packet is buffered for extended amount of time. On the other hand, as soon as the NIC receives a packet that it classifies as belonging to an active communication the *Long Idle* mode is set to off and these packets are pushed to the platform instantly.

#### **5. Evaluation and analysis**

In our analysis we evaluate the performance of the *Long idle* technology. First we quantify its traffic classification accuracy and then we evaluate the power and performance impact in case of both idle and active workloads with and without using *Long idle*.

#### **5.1 Traffic categorization accuracy**

10 Will-be-set-by-IN-TECH

destination, interarrival times, and number of transmitted and received packets are collected and based on the conditions outlined above the ongoing communication is classified as active versus background one. Figure 9 depicts the wireless NIC sleep time update based on the classification of the ongoing traffic. If *Long Idle* is detected the Wireless NIC sleeps longer and the received packet is buffered for extended amount of time. On the other hand, as soon as the NIC receives a packet that it classifies as belonging to an active communication the *Long*

In our analysis we evaluate the performance of the *Long idle* technology. First we quantify its traffic classification accuracy and then we evaluate the power and performance impact in case

*Idle* mode is set to off and these packets are pushed to the platform instantly.

of both idle and active workloads with and without using *Long idle*.

Fig. 9. Long Idle WNIC Implementation

**5. Evaluation and analysis**

*Long Idle* relies on the NIC's ability to categorize the network traffic based on the conditions and algorithms outlined in Section 4. We use real-network traffic traces (as mentioned in Section 3) that represent hours of idle and active platform network activities as an input to our *Long idle* simulator running the classification and traffic-type-specific data handling algorithms to quantify the accuracy of these algorithms.

(b) Wireless Environment

#### Fig. 10. Long Idle Classification Algorithm Accuracy

Figure 10 analyzes the accuracy of the traffic classification algorithm in two environments wired home environment and wireless enterprise environment. The trend of this result is the same for all other tested networks. The false positive ratio represents the percentage of time idle traffic has been classified as active (i.e., the platform is waken up unnecessarily to process these idle packets) to the total time until it is correctly classified as idle. The false negative ratio represents the percentage of time active traffic is classified as idle (i.e., packets are held in the NIC and processing is delayed).

Figure 10 illustrates the effectiveness of our traffic categorization algorithm. As shown in the figure the average accuracy of the algorithm at a sampling window size of 1 sec in an enterprise environment is about 92% and in a home environment is close to 96%. The false negative percentage is very small which indicates that active traffic is never classified as idle, as a result active traffic is not buffered in the NIC and hence the active traffic performance and the user experience are unaffected. As shown in the figure most of the classification errors are

Input Rate 802.11 Observed Rate Long Idle Observed Rate 256 Kbps 256 Kbps 256 Kbps 512 Kbps 512 Kbps 512 Kbps 1 Mbps 0.995 Mbps 0.985 Mbps 2 Mbps 1.997 Mbps 1.964 Mbps 10 Mbps 9.875 Mbps 9.7656 Mbps

<sup>83</sup> Energy Efficiency of

not degraded when *Long Idle* is used which indicates that the NIC correctly classified the FTP session as an active communication session and backed off from buffering any network packets. The slight difference in throughput between 802.11 PSM and that achieved by *Long Idle* is because *long Idle* has a slower responsiveness, as mentioned in Section 5.1, and the "first" packet in an active communication is delayed until at least one sampling window (i.e.,

> VoIP Session 802.11 PSM Quality Long Idle Quality Two Way 4.37 4.34

VoIP MOS is a numerical method of expressing voice and video quality it is a measurable indication of the perceived quality of the media received after being transmitted and eventually compressed using codecs. A MOS value larger than 4.0 is what VOIP services targets as a good quality VoIP session *ITU-T Rec. G.729 Annex B, A silence compression scheme for G.729 optimized for terminals conforming to Rec. V.70* (1996). Table 2 represents the MOS results achieved by the NIC with and without *Long Idle*. Similarly, when *Long Idle* is enabled it correctly identifies the active communication and the packets are not buffered in NIC and hence no user experience impact. Similar to the FTP case, the small difference in the MOS

Typical platform power management relies on guiding individual platform components or the whole platform into low energy *(sleep)* states when the platform has been observed "idle" for some time with no active workloads. Individual power management techniques differ in how deep the sleep state is, the algorithms used to enter and exit the sleep states and optimizations to extend these sleep states as long as possible. In this chapter, we showed that significant energy is wasted at the platform level while it is idle and connected to a network because the system is processing background and management traffic. We showed that these background packets arrive at a rate high enough that can prevent the system from taking full advantage of the platform-level power management states. We quantified the negative impact of network

We proposed *Long Idle* a low overhead technique that is implemented inside the network interface card with no dependency from the host, application or the network infrastructure to classify the ongoing traffic into active versus background without deep packet inspection and to locally buffer the background traffic in the NIC to guarantee uninterrupted sleep periods for the platform. We showed that when our scheme is used the total platform sleep time increases by up to 40% with no performance degradation and no user experience impact.

One Way 4.37 4.33

4.37 4.37

Table 1. FTP Throughput with Long Idle

Connected Mobile Platforms in Presence of Background Traffic

Table 2. VoIP Quality with Long Idle

**6. Conclusion**

1 sec) has elapsed before the ongoing traffic is classified as active.

score is attributed to the delay experienced by the first packet.

connectivity on platform energy-efficiency and showed.

false positives (idle classified as active) which indicates that a small percentage of the power reduction opportunities have been missed and the platform is waken up unnecessarily.We confirm these simulations with implementation results presented in the next subsections.As expected the classification accuracy improves with increasing the sampling window but, on the other hand, the response time defined as how fast the platform will react to traffic changes degrades. As shown in the figure the classification accuracy saturates mostly at a sampling window size of 700 msec. Based on our results a sampling window size of 1 sec achieves a good classification accuracy and good responsiveness with unaffected user experience.

#### **5.2 Idle traffic impact**

We implemented *Long Idle* classification algorithm in a prototype Intel WiFi 5350 NIC *Intel WiFi Link 5100 Series Specifications* (2008). In our implementation we used a sampling window size of 1 sec and NIC idle traffic buffering timeout of 300 msec.

Fig. 11. Implementation Impact on Idle Traffic

Figure 11 plots the behavior of the NIC with idle traffic. Whenever the platform is not engaged in an active communication *Long Idle* technology will guarantee a quietness period as large as the buffering timeout value (e.g., 300 msec) set in the NIC and therefore the platform will stay in a low-power state uninterrupted during this time. Compared to Figure 5(b), energy gain achieved by Long Idle is about 23% for wireless environment and about 41% for the wired environment. It should be noted that we assumed that *τidle* = 0 (see Section 3) to report worst case benefit and any value for *τidle >* 0 will increase the energy gain of *Long Idle* even further.

#### **5.3 Active traffic impact**

To quantify the effect of *Long Idle* on active workloads we used our NIC prototype implementation to run various NetIQ Chariot *NetIQ Chariot 4.0 testing tools* (2001) benchmarks. NetIQ Chariot is a prepackaged real-world benchmarking tool widely used for network performance evaluation. We present results for FTP throughput and VoIP Mean Opinion Score (MOS) tests.

Table 1 represents the network throughput achieved by the NIC with and without *Long Idle* compared to the traditional 80.11 PSM. As shown in the table, the network throughout is


Table 1. FTP Throughput with Long Idle

12 Will-be-set-by-IN-TECH

false positives (idle classified as active) which indicates that a small percentage of the power reduction opportunities have been missed and the platform is waken up unnecessarily.We confirm these simulations with implementation results presented in the next subsections.As expected the classification accuracy improves with increasing the sampling window but, on the other hand, the response time defined as how fast the platform will react to traffic changes degrades. As shown in the figure the classification accuracy saturates mostly at a sampling window size of 700 msec. Based on our results a sampling window size of 1 sec achieves a good classification accuracy and good responsiveness with unaffected user experience.

We implemented *Long Idle* classification algorithm in a prototype Intel WiFi 5350 NIC *Intel WiFi Link 5100 Series Specifications* (2008). In our implementation we used a sampling window

Figure 11 plots the behavior of the NIC with idle traffic. Whenever the platform is not engaged in an active communication *Long Idle* technology will guarantee a quietness period as large as the buffering timeout value (e.g., 300 msec) set in the NIC and therefore the platform will stay in a low-power state uninterrupted during this time. Compared to Figure 5(b), energy gain achieved by Long Idle is about 23% for wireless environment and about 41% for the wired environment. It should be noted that we assumed that *τidle* = 0 (see Section 3) to report worst case benefit and any value for *τidle >* 0 will increase the energy gain of *Long Idle* even further.

To quantify the effect of *Long Idle* on active workloads we used our NIC prototype implementation to run various NetIQ Chariot *NetIQ Chariot 4.0 testing tools* (2001) benchmarks. NetIQ Chariot is a prepackaged real-world benchmarking tool widely used for network performance evaluation. We present results for FTP throughput and VoIP Mean

Table 1 represents the network throughput achieved by the NIC with and without *Long Idle* compared to the traditional 80.11 PSM. As shown in the table, the network throughout is

size of 1 sec and NIC idle traffic buffering timeout of 300 msec.

Fig. 11. Implementation Impact on Idle Traffic

**5.3 Active traffic impact**

Opinion Score (MOS) tests.

**5.2 Idle traffic impact**

not degraded when *Long Idle* is used which indicates that the NIC correctly classified the FTP session as an active communication session and backed off from buffering any network packets. The slight difference in throughput between 802.11 PSM and that achieved by *Long Idle* is because *long Idle* has a slower responsiveness, as mentioned in Section 5.1, and the "first" packet in an active communication is delayed until at least one sampling window (i.e., 1 sec) has elapsed before the ongoing traffic is classified as active.


Table 2. VoIP Quality with Long Idle

VoIP MOS is a numerical method of expressing voice and video quality it is a measurable indication of the perceived quality of the media received after being transmitted and eventually compressed using codecs. A MOS value larger than 4.0 is what VOIP services targets as a good quality VoIP session *ITU-T Rec. G.729 Annex B, A silence compression scheme for G.729 optimized for terminals conforming to Rec. V.70* (1996). Table 2 represents the MOS results achieved by the NIC with and without *Long Idle*. Similarly, when *Long Idle* is enabled it correctly identifies the active communication and the packets are not buffered in NIC and hence no user experience impact. Similar to the FTP case, the small difference in the MOS score is attributed to the delay experienced by the first packet.

#### **6. Conclusion**

Typical platform power management relies on guiding individual platform components or the whole platform into low energy *(sleep)* states when the platform has been observed "idle" for some time with no active workloads. Individual power management techniques differ in how deep the sleep state is, the algorithms used to enter and exit the sleep states and optimizations to extend these sleep states as long as possible. In this chapter, we showed that significant energy is wasted at the platform level while it is idle and connected to a network because the system is processing background and management traffic. We showed that these background packets arrive at a rate high enough that can prevent the system from taking full advantage of the platform-level power management states. We quantified the negative impact of network connectivity on platform energy-efficiency and showed.

We proposed *Long Idle* a low overhead technique that is implemented inside the network interface card with no dependency from the host, application or the network infrastructure to classify the ongoing traffic into active versus background without deep packet inspection and to locally buffer the background traffic in the NIC to guarantee uninterrupted sleep periods for the platform. We showed that when our scheme is used the total platform sleep time increases by up to 40% with no performance degradation and no user experience impact.

**5** 

**The Energy Efficient Techniques in the DCF of** 

**802.11 and DRX Mechanism of LTE-A Networks** 

Mobile devices such as notebooks and smart phones have replaced the personal computer as main personal information devices. These devices drive a strong demand for wireless networks and wireless communication for the rapidly growing number of Internet users. Since mobile terminals will be severely constrained by their limited battery endurance, it is essential that new protocols and control mechanisms based on the existing 802.11 standard [45] [46] [47] [48] and Long Time Evolution-Advanced (LTE-A) networks [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] be devised to reduce power consumption. It should be very energy efficient in 802.11 networks due to its short transmission and reception distance between stations and Access Point (AP). The characteristic of DCF is distributed, so every active station must turn on its transceiver to listen to the common share channel. It is apparent that the cost of DCF is very low due to this distributed characteristic. However, the energy efficiency of DCF is very poor due to this distributed characteristic also. Every active station must wait on this common shared channel by turning on its transceiver. It is socalled idle listening. Researchers have investigated a wide variety of techniques to limit the power consumption of the wireless network interface. Feeney and Nilsson [61] showed that an Orinoco Silver 802.11 card consumes on average, 47.4 *mW* with the receiver turned off (sleeping), 739.4 *mW* while listening to an idle channel, 900.6 *mW* while receiving data and 1346.2 *mW* while transmitting. Their results show that, unnecessary transmissions are costly, so is leaving the receiver on when it is receiving nothing. Hence, reducing the idle listening time is important for reducing the power consumption of 802.11. In the latter sections, we will illustrate that the overheads on idle listening to transmit an uplink data frame in the 802.11 networks are really extremely large especially when the number of active stations is large or the data frame is extremely short. Therefore the technique to conserve energy consumption is really very important. Fortunately the LTE and LTE-A propose the Discontinuous Reception (DRX) to conserve the energy consumption in idle listening. The DRX cycle, on-duration and sleep duration are scheduled by eNB. The UE just wakes up in the on-duration to receive possible PDCCH. If no further frame reception is indicated by PDCCH or no PDCCH is received, the user equipment (UE) can enter sleep mode in the

**1. Introduction** 

Kuo-Chang Ting, Hwang-Cheng Wang,

*Minghsin University of Science and Technology* 

*National Ilan University National Chi Nan University* 

*Taiwan* 

Fang-Chang Kuo, Chih-Cheng Tseng and Ping Ho Ting

#### **7. References**

