Article

Opportunistic Hybrid Transport Protocol (OHTP) for Cognitive Radio Ad Hoc Sensor Networks Yousaf Bin Zikria 1 , Summera Nosheen 2 , Farruh Ishmanov 3 and Sung Won Kim 1, * Received: 29 September 2015; Accepted: 9 December 2015; Published: 15 December 2015 Academic Editor: Leonhard M. Reindl 1 2 3

*

Department of Information and Communication Engineering, Yeungnam University, 280, Daehak-Ro, Gyeongsan, Geongbuk 38541, Korea; [email protected] Comsats Institute of Information Technology, Park Rd, Islamabad 45550, Pakistan; [email protected] Department of Electronics and Communication Engineering, Kwangwoon University, 447-1 Wolgye-dong, Nowon-gu, Seoul 01897, Korea; [email protected] Correspondence: [email protected]; Tel.: +82-53-810-2483; Fax: +82-53-810-4742

Abstract: The inefficient assignment of spectrum for different communications purposes, plus technology enhancements and ever-increasing usage of wireless technology is causing spectrum scarcity. To address this issue, one of the proposed solutions in the literature is to access the spectrum dynamically or opportunistically. Therefore, the concept of cognitive radio appeared, which opens up a new research paradigm. There is extensive research on the physical, medium access control and network layers. The impact of the transport layer on the performance of cognitive radio ad hoc sensor networks is still unknown/unexplored. The Internet’s de facto transport protocol is not well suited to wireless networks because of its congestion control mechanism. We propose an opportunistic hybrid transport protocol for cognitive radio ad hoc sensor networks. We developed a new congestion control mechanism to differentiate true congestion from interruption loss. After such detection and differentiation, we propose methods to handle them opportunistically. There are several benefits to window- and rate-based protocols. To exploit the benefits of both in order to enhance overall system performance, we propose a hybrid transport protocol. We empirically calculate the optimal threshold value to switch between window- and rate-based mechanisms. We then compare our proposed transport protocol to Transmission Control Protocol (TCP)-friendly rate control, TCP-friendly rate control for cognitive radio, and TCP-friendly window-based control. We ran an extensive set of simulations in Network Simulator 2. The results indicate that the proposed transport protocol performs better than all the others. Keywords: transport protocol; congestion control

cognitive radio ad hoc sensor network;

opportunistic;

1. Introduction Extensive enhancements in wireless technologies and ever-increasing bandwidth-intensive applications require more spectrum resources. The current spectrum allocation policy gives full rights only to licensed operators. However, this is one of the factors contributing to spectrum underutilization. A Federal Communications Commission (FCC) survey identified the inefficient usage of assigned spectrum [1]. To solve this issue, the cognitive radio network (CRN) [2] was introduced to exploit the holes in spectrum bands. A CRN is a viable solution to improve the efficiency of spectrum usage and network capacity. There are two types of users in a CRN: the licensed, or primary, users (PUs) and the unlicensed, or secondary, users (SUs). The SUs opportunistically access the spectrum band when the PUs are not using it. In a CRN, the primary Sensors 2015, 15, 31672–31686; doi:10.3390/s151229871

www.mdpi.com/journal/sensors

Sensors 2015, 15, 31672–31686

constraint is to protect PU transmissions. Therefore, when a PU arrives, any SU must vacate the channel immediately. The CRN may consist of a single channel or be multi-channel. With a single channel, SUs have to wait for availability of the channel for communications. In multi-channel scenarios, the SUs still have to leave a channel upon PU arrival. However, if another channel is available, they can switch to that channel and resume communications. Otherwise, SUs wait until another channel is free. The cognitive radio ad hoc network (CRAHN) [3] does not have any infrastructure backbone. Cognitive radio (CR) users communicate with other CR users in an ad hoc manner on both licensed and unlicensed spectrum bands. In a CRAHN, each user needs to have all the CR capabilities and is responsible for all communications actions based on local observation. The cognitive radio ad hoc sensor network (CRASN) [4] is a specialized ad hoc network of distributed wireless sensors that are equipped with cognitive radio capabilities. A wireless node selects a vacant channel for communications and leaves the channel upon the arrival of a licensed user. Cognitive radio is one of the most promising techniques to enhance the efficiency of wireless sensor networks (WSNs). A CRASN increases spectrum utilization, increases network efficiency, and extends the lifetime of the WSN [5]. In this paper, we propose a new transport protocol named opportunistic hybrid transport protocol for cognitive radio ad hoc sensor networks (OHTP-CRASN). The research objective is to provide a new congestion control mechanism by considering the dynamic nature of cognitive radio ad hoc sensor networks. Furthermore, it takes advantage of window-based and rate-based protocols according to the network conditions to improve overall system performance. The rest of the paper is organized as follows. Section 2 discusses the related work. In Section 3, we explain our proposed opportunistic hybrid transport protocol for cognitive radio ad hoc sensor networks. Section 4 provides the details of the simulation environment and discusses the results. Finally, Section 5 concludes the paper. 2. Related Work Transmission Control Protocol (TCP) [6] is the de facto transport layer connection-oriented and reliable end-to-end protocol. It provides reliable data delivery with the help of flow control and congestion control mechanisms. TCP is the window-based protocol that uses sequence numbers and acknowledgements (ACKs) to attain reliability. The sender and receiver exchange data associated with segments. When the receiver receives a segment, it sends an acknowledgement to the sender indicating correct reception of the segment. On successful reception of an ACK, the sender transmits the next segment. If the sender gets two ACKs for the same segment, it is called a duplicate ACK. On reception of three duplicate ACKs, the sender assumes that segment and its data were lost and retransmits. Furthermore, TCP also employs a timeout mechanism to detect losses. It starts a timer immediately after transmitting a segment and waits for the timeout event. If it gets an ACK before the timer expires, it considers the data successfully delivered. Otherwise, on the occurrence of a timeout, it considers the segment and its data as lost and retransmits the same segment. The timeout event forces TCP to initiate the slow start algorithm. The timeout interval is called retransmission timeout (RTO) [7]. TCP adjusts its window according to the network conditions by using an additive-increase multiplicative-decrease (AIMD) [6,8] strategy. TCP starts the window with one packet or some large value. Moreover, the window is increased exponentially by one segment for every successive ACK until the source reaches network capacity. This phase is called slow start (ss), and the capacity estimate is called the ss threshold (ssthresh). Hereafter, congestion avoidance (CA) starts, and the window is increased by one segment for every round-trip time (RTT) until loss is detected. There are two methods to detect packet loss, as explained above: RTO expiration and three duplicate ACKs. In the case of loss, TCP considers the cause to be network congestion and reduces the current congestion window to half. Network congestion means that the receiver’s network resources are overloaded due to the

31673

Sensors 2015, 15, 31672–31686

current sender rate. Hence, packet loss and delay increase, and the network can collapse in the Sensors 2015, 15, page–page worst-case scenario. To cope with this critical issue, many variants of TCP, like Tahoe [6], Reno [9], New Reno [10], SACK [11], Vegas [12], etc., offer different solutions to handle congestion control cope with this critical issue, many variants of TCP, like Tahoe [6], Reno [9], New Reno [10], SACK [11], and avoidance issues. TCP estimates round-trip time using the SYN estimator. The sender sends Vegas [12], etc., offer different solutions to handle congestion control and avoidance issues. TCP the SYN segment to the receiver, and the receiver sends back a SYN-ACK on successfully receiving estimates round-trip time using the SYN estimator. The sender sends the SYN segment to the receiver, the segment. Then, the sender to on thesuccessfully SYN-ACK receiving segment the with an ACK segment. The time and the receiver sends back responds a SYN-ACK segment. Then, the sender interval between the arrival of a SYN segment and the arrival of an ACK segment is called the responds to the SYN-ACK segment with an ACK segment. The time interval between the arrival of a RTT of theSYN connection. RTT minimum total transmission delay. The RTT segment and theindicates arrival of the an ACK segment is called the RTT ofand the propagation connection. RTT indicates the minimum total transmission estimator is illustrated in Figure 1.and propagation delay. The RTT estimator is illustrated in Figure 1.

Figure 1. TCP RTT sample calculation.

Figure 1. TCP RTT sample calculation.

TCP CRAHN [13] is the variant of TCP and based on the classical TCP NewReno. They added TCPfunctionality CRAHN [13] is the variant of TCP and based on the However, classical TCP They added the of channel switching and periodic switching. TCP NewReno. CRAHN congestion control is completely the same as TCP. Theperiodic functionality of periodic sensingTCP is achieved using the the functionality of channel switching and switching. However, CRAHN congestion extra to the theTCP. routing This approach is not scalable as is theachieved nodes have to the control is message completely thenodes sameinas Thepath. functionality of periodic sensing using accommodate multiple flows from different TCP sources. Due to the additional states and control extra message to the nodes in the routing path. This approach is not scalable as the nodes have to messages, it is difficult to implement and requires additional buffer space along the path. Moreover, accommodate multiple flows from different TCP sources. Due to the additional states and control it requires feedback information from relay nodes and cross-layer collaboration from lower layers. messages, it is difficult to implement and requires additional buffer space along the path. Moreover, Hence, the end-to-end principle is broken [14]. it requires feedback relay andofcross-layer collaboration from lower layers. RETP [15] isinformation proposed tofrom prolong thenodes lifetime nodes by energy efficiency. Every node Hence, the end-to-end principle is broken [14]. determines its operating channel in a distributed manner although the scheduling of sending data is RETP [15] centrally is proposed to prolong theassumed lifetimethat of every nodesnode by can energy Every determined by a sink node. They reachefficiency. the sink in one hop. node determines in abedistributed manner although the scheduling of sending data Hence, ititsis operating not scalablechannel and cannot used in multihop environment. In contrast to window-based mechanisms, rate-based control to hop. is determined centrally by a sink node.congestion They assumed that every node can reachwas theproposed sink in one address congestion handling under TCP. TCP-friendly rate control (TFRC) [16] is the first rate-based Hence, it is not scalable and cannot be used in multihop environment. transport protocol. These authors congestion proposed increasing the sending rate slowly in response to a In contrast to window-based mechanisms, rate-based control was proposed to decrease in the loss event rate. It does not halve the sending rate in response to a single loss event. address congestion handling under TCP. TCP-friendly rate control (TFRC) [16] is the first rate-based However, in the case of several consecutive losses, it halves the sending rate. To adjust the sending transport protocol. These authors proposed increasing the sending rate slowly in response to a rate, the receiver informs the sender as to the reception of packets per round-trip time. The sender will decrease inthe thesending loss event does not halve thefor sending rate round-trip in response to aThe single reduce rate ifrate. it hasItnot received feedback consecutive times. loss loss eventevent. However, the caseatof consecutive losses, it halves sendingThereafter, rate. To adjust thewill sending rate is in calculated theseveral receiver, and the receiver provides it to the the sender. the sender rate, adjust the receiver informs sender asIttouses the the reception packets per round-trip time. The the sending ratethe accordingly. methodofcalled Average Loss Interval (ALI) to sender obtain will reduce sending rate if rate. it hasThis notsmoothing received feedback consecutive round-trip times. to The loss event thethe smoothed sending is done in for order to circumvent violent reaction a single unexpected rate is calculatedloss at event. the receiver, and the receiver provides it to the sender. Thereafter, the sender will are three factors determining throughput: TCP-friendly loss eventadjust theThere sending rate main accordingly. It uses theTFRC method called Average Lossequations, Interval (ALI) to obtain rate estimation, and network delays. The sender calculates a new RTT sample whenever a feedback the smoothed sending rate. This smoothing is done in order to circumvent violent reaction to a single packet is received. Figure 2 illustrates the TFRC RTT sample calculation. When a feedback packet is unexpected loss event. received, the sender performs the following five steps: (i) calculate the most recent sample of the RTT;

There are three main factors determining TFRC throughput: TCP-friendly equations, loss event-rate estimation, and network delays. The sender calculates a new RTT sample whenever a feedback packet is received. Figure 2 illustrates the 3 TFRC RTT sample calculation. When a feedback 31674

Sensors 2015, 15, 31672–31686

packet is received, the sender performs the following five steps: (i) calculate the most recent sample of theSensors RTT;2015, (ii)15,estimate page–pagea new smoothed RTT; (iii) calculate the TCP retransmission timeout value; Sensors 2015, 15, page–page (iv) adjust the sending rate; and (v) reset the non-feedback timer. (ii) estimate a new smoothed RTT; (iii) calculate retransmission timeout value; (iv) adjust the TFRC ensures a smooth transmit rate usingthe theTCP following function: (ii) estimate new(v) smoothed (iii) calculate the TCP retransmission timeout value; (iv) adjust the sending rate;a and reset theRTT; non-feedback timer. sending rate; and (v) reset thetransmit non-feedback timer. s TFRC ensures a smooth rate using the function: ˜ following T “ transmit c rate using the c ¸ function: TFRC ensures a smooth following ` ˘ 2p 3p R= ` ` tRTO . 3 .p. 1 ` 32p2 3 8 (1) = + + . . . + (1) + + . . . +

(1)

Sending raterate T isTaisfunction ofofthe theround-trip round-trip time TCP retransmit timeout Sending a function thepacket packet size, size, the time R, R, thethe TCP retransmit timeout Sending rate T is a function of the packet size, the round-trip time R, the TCP retransmit timeout tRTO ,tand loss event rate p. Sender and receiver both have to measure the packet size, round-trip time, RTO, and loss event rate p. Sender and receiver both have to measure the packet size, round-trip time, t RTO , and loss event rate p. Sender and receiver both have to measure the packet size, round-trip time, and loss rate.rate. tRTO is isset approximation,and and this is reasonable for acquiring and event loss event tRTO settoto44ˆ× R R for approximation, this is reasonable for acquiring TCP TCP and loss[17]. event rate. tRTO set to 4 ×TFRC R for congestion approximation, andmechanism. this is reasonable for acquiring TCP friendliness Figure 3 shows the control mechanism. friendliness [17]. Figure 3isshows the TFRC congestion control friendliness [17]. Figure 3 shows the TFRC congestion control mechanism.

Figure 2. TFRC RTT sample calculation.

Figure samplecalculation. calculation. Figure2.2.TFRC TFRC RTT RTT sample

Figure 3. TFRC congestion control mechanism. Figure3.3.TFRC TFRCcongestion congestion control Figure controlmechanism. mechanism. 4 4

31675

Sensors 2015, 15, 31672–31686

Sensors 2015, 15, page–page

In order to utilize bandwidth more efficiently, TCP-Friendly Rate Control for Cognitive Radio In order utilize bandwidth moreinstant efficiently, TCP-Friendly Rate forspectrum-related Cognitive Radio (TFRC-CR) [18]towas proposed. It allows changes in the sender rateControl based on (TFRC-CR) [18] was adjusts proposed. allows instant in theand sender rate based on spectrum-related changes. TFRC-CR theItsending rate bychanges identifying differentiating between spectrum changes. TFRC-CR adjusts in the rate by identifying and differentiating between spectrum change and true congestion thesending network. Furthermore, it provides the time to restart transmissions change and true congestion in the network.enhances Furthermore, it provides thefits time to restart transmissions after a pause upon PU arrival. TFRC-CR overall TFRC and into the CRN paradigm. after a pause upon PUresponds arrival. TFRC-CR overall TFRC in and into the rate. CRNThis paradigm. Despite this, TFRC-CR slowly to enhances an immediate decrease thefits loss event is due Despite responds to an immediate in thehistory. loss event rate. This in is due to a flaw this, in theTFRC-CR computation of theslowly loss event rate that reliesdecrease on loss event Additionally, the to a flaw in the computation of the loss event rate that relies on loss event history. Additionally, in case of a spectrum change to an idle spectrum with larger capacity, it takes a longer time to adjust the the case ofnewly a spectrum change to an idle spectrum capacity, it takes a longer time to adjust rate to the available capacity. Thus, it affectswith RTTlarger and RTO values. the rate to the newly available Thus, it affects RTT and RTO values. Some researchers [19,20] capacity. suggested sender-based measurement of RTT because most end researcherscapabilities. [19,20] suggested sender-based of RTT because mostavoidance end users usersSome lack processing In addition, TFRC measurement lacks the fine-grained congestion lack processing capabilities. In addition, TFRC lacks the fine-grained avoidance mechanism that TCP ACK-clocking provides. Therefore, a new mechanismcongestion called TCP-friendly mechanism that TCP (TFWC) ACK-clocking new mechanism called TCP-friendly window-based control [21] wasprovides. proposed.Therefore, It modifiesa the TFRC throughput model according window-based control (TFWC) [21] was proposed. It modifies the TFRC throughput model according to Equation (2) to fit a window-based mechanism. to Equation (2) to fit a window-based mechanism. 1 ˜ c ¸ (2) W“ c = 2p ` 12 3p .p. `1 ` 32p2 ˘ (2) 3 8

+

. .

+

where W is the window size in the packets, and p is the loss event rate. The sender computes the where loss W isand the calculates window size in the packets, and (cwnd) is theusing loss event rate. equation. The senderIf computes the packet the congestion window the TFWC the sequence packet loss and calculates the congestion (cwnd) using theACK-clock TFWC equation. If the sequence number of the new data waiting to be sent window fulfills Equation (3), the is generated. number of the new data waiting to be sent fulfills Equation (3), the ACK-clock is generated. sequence number of data ď ≤ cwnd ` +unacknowleged sequence number new

(3)(3)

The TFWC TFWC cwnd cwnd mechanism mechanism isis presented presented inin Figure Figure4.4. The TFWC receiver receiver provides provides an an The acknowledgement vector (AckVec) and time-stamp echo to the sender to calculate the cwnd and RTT. acknowledgement vector (AckVec) and time-stamp echo to the sender to calculate the cwnd and RTT. the ACK ACK flow flow isis disrupted, disrupted, the the TFWC TFWC smoothness smoothnessisisreduced. reduced. This This protocol protocol is is proposed proposed for for wired wired IfIf the networks, and and therefore, therefore, cannot cannot handle handle the the challenges challenges related related to to aa CRN, CRN, such such as as frequent frequent spectrum spectrum networks, changesdue dueto toPU PUactivity. activity.Hence, Hence,the thelosses lossesdue dueto tothe thedynamic dynamicnature natureof ofthe thenetwork networkare areconsidered considered changes congestion. As As aa result, result, itit adversely adverselyaffects affectsthe thesending sendingrate rateand andunderutilizes underutilizesnetwork networkcapacity. capacity. congestion.

Figure4. 4. TFWC TFWC congestion congestionwindow windowmechanism. mechanism. Figure

3. OHTP-CRASN 3. OHTP-CRASN The paradigm paradigmof ofthe thetransport transportprotocol protocolchanges changesin inaaCRASN CRASNand andneeds needsto toadapt adaptto to the the dynamic dynamic The nature of of the the CRN. CRN. The The first first and and foremost foremost important importantthing thing isis to to detect detect the the PU PU during during transmissions. transmissions. nature Afterwards, losses during the handover from one channel to a vacant channel should not be considered congestion. In fact, these losses are interruption losses [22]. If it is not tackled carefully, 31676 5

Sensors 2015, 15, 31672–31686

Afterwards, losses during the handover from one channel to a vacant channel should not be considered congestion. Sensors 2015, 15, page–page In fact, these losses are interruption losses [22]. If it is not tackled carefully, the performance of the transport protocol is greatly hampered. Hence, in an efficient transport the performance of the between transportfake protocol is greatly hampered. Hence,We in use an efficient transport protocol, differentiating and true congestion is essential. the FCC-specified protocol, differentiating fake true congestion is essential. We use instant the FCC-specified database and connectivitybetween directions [23]and to detect the PU appearance at a specific in time on a database and connectivity directions [23] to detectdatabases the PU appearance at a specific instantprotocol in time on tagged channel. The advantage of FCC-mandated is that it allows the transport to aintegrate tagged channel. The advantage of FCC-mandated databases is that it allows the transport with a designated spectrum database. It contains information about PU’s location, protocol channel, to integrate with a and designated It need contains information aboutthePU’s location, transmission power activity spectrum over time. database. There is no for the feedback from intermediate channel, powerThe anddatabase activity over time.only There is no for the feedbackthe from the nodes or transmission underlying layers. is polled when it isneed needed. Whenever packet intermediate nodes or underlying layers. onlyitwhen it is needed. Whenever loss is deteced, it immediately checks the The PU’sdatabase activity. is Wepolled consider opportunistic because all the the packetareloss is deteced, itand immediately PU’s activity. We consider it opportunistic channels homogeneous, shifting to checks the newthe channel while retaining the same sending rate because all theanchannels are homogeneous, and shifting to thetime. new channel while retaining the same may provide opportunity to deliver more data in a given Hence, exploring the window of sending rate will mayenhance provide system an opportunity to deliver more data in a given time. Hence, exploring the opportunity throughput. window of opportunity will enhance system throughput. Figure 5 portrays OHTP congestion control mechanism. OHTP is the sender-driven congestion Figure 5 portrays OHTP congestion controlrate mechanism. OHTP the sender-driven congestion control protocol. It calculates the transmission with the help of is receiver feedback. The receiver control protocol. It calculates the transmission rate with the help of receiver feedback. The receiver reports feedback at least once per round-trip time. The feedback contains the number of packets reports at least percalculates round-trip time. Theevery feedback contains number of packets receivedfeedback by receiver. Theonce source cwnd upon feedback. In the case of packet loss, received by immediately receiver. The checks source calculates cwnd upon on every feedback. In the case the of packet loss, the the sender for any PU activity specific channel using FCC-specified sender immediately checks itforreaffirms any PU activity on specificloss channel using the FCC-specified database. database. If PU is active, the interruption and stores cwnd. After moving to If PU is active, it reaffirms stores the cwnd.channel. After moving to the the next available channel,the it interruption resumes the loss rateand from the previous In case the next loss available is due to channel, resumesitthe rate fromthe thecwnd previous In case the loss is due to erroneous channel, erroneousit channel, determines usingchannel. TCP throughput equation. In addition, it reduces the it determines the cwnd using TCPevent throughput equation. In addition, itrate reduces the cwnd cwnd by packet loss ratio. In the of no packet loss, transmission increases until by thepacket upper loss In the event of no packet loss, transmission rate increases until the upper limit is reached. limitratio. is reached.

Figure Figure 5. 5. OHTP OHTP congestion congestion control control mechanism. mechanism.

Figure 6 depicts the OHTP timing diagram. Here, the data represent the packets transmitted and feedback means the list of packets successfully received by receiver. By checking the feedback, the sender knows the lost packets. The sender looks into the feedback for any missing packets excluding the last three packets by applying TCP three 31677 duplicate rule. When the packet loss is detected, it determines whether it is the interruption loss or the normal loss. If it is the interruption loss, it keeps 6

Sensors 2015, 15, 31672–31686

Figure 6 depicts the OHTP timing diagram. Here, the data represent the packets transmitted and feedback means the list of packets successfully received by receiver. By checking the feedback, the sender knows the lost packets. The sender looks into the feedback for any missing packets excluding the last three packets by applying TCP three duplicate rule. When the packet loss is Sensors 2015, 15, page–page detected, it determines Sensors 2015, 15, page–pagewhether it is the interruption loss or the normal loss. If it is the interruption loss,the it current keeps the current cwndtransmission. for further Otherwise, transmission. Otherwise, it calculates using the cwnd for further it calculates the cwnd using thethe TCPcwnd throughput current cwnd for further transmission. Otherwise, it calculates the cwnd using the TCP throughput TCPthe throughput equation and additionally reduces the cwnd by packet loss ratio. The OHTP finite equation and additionally reduces the cwnd by packet loss ratio. The OHTP finite state machine (FSM) equation and additionally reduces the cwnd by packet loss ratio. The OHTP finite state machine (FSM) stateismachine shown in(FSM) Figureis7.shown in Figure 7. is shown in Figure 7.

Figure 6. OHTP timing diagram.

Figure diagram. Figure6.6.OHTP OHTP timing timing diagram.

Figure 7. OHTP FSM.

Figure FSM. Figure 7. 7. OHTP OHTP FSM.

3.1. Hybrid State

Hybrid State 3.1. 3.1. Hybrid State

This is the default state of the OHTP-CRASN. The protocol returns to this state after interruption. This is the default state ofthe theOHTP-CRASN. OHTP-CRASN. The protocol toto this state after interruption. This is the default state The protocol returns this state after interruption. The protocol starts with a of linear increase of cwnd until a packetreturns loss event occurs. On the occurrence The protocol starts with a linear increase of cwnd until a packet loss event occurs. On the Theof protocol startsitwith a linear increase cwnd until a packet loss is event On the occurrence of packet loss, instantly enters the PUofdetection state. This state also occurs. responsible for occurrence calculating of it instantly enters the detection PU state. This is also responsible forcalculating calculating the packet loss, itloss, instantly enters the PU state. This state is also responsible for thepacket sending rate for packets in bytes perdetection second based on thestate decision from the window/rate-based the sending rate of forpacket packets indue bytes per second the decision from the window/rate-based state. In the loss to fading or anbased erroneous channel, the transmission rate is directly sending rate forcase packets in bytes per second based on theondecision from the window/rate-based state. state. In the case of packet loss due to fading or an erroneous channel, the transmission ratein is directly calculated the loss help due of a TCP throughput adjust the rate or In the case ofwith packet to fading or an equation. erroneousWe channel, thetransmission transmission rate bytes is directly calculated the at help of a TCP throughput We packet adjust the transmission ratethe in window bytes or packets with bywith looking the interval history. Ifequation. there is 10% loss,the then we reduce calculated the help ofloss a TCP throughput equation. We adjust transmission rate in bytes packets by looking at the loss packet intervalloss history. If there is 10% packet loss, then we reduce the window by the same ratio. However, due to PU appearance is not treated as channel congestion. or packets by looking at the loss interval history. If there is 10% packet loss, then we reduce the by the same ratio. However, packet loss due to PU appearance is not treated as channel congestion. The transmission rate is not changed. This is the opportunistic approach, which increases throughput The transmission rate is not changed. This is the opportunistic approach, which increases throughput aggressively. 31678 aggressively. 7 7

Sensors 2015, 15, 31672–31686

window by the same ratio. However, packet loss due to PU appearance is not treated as channel congestion. The transmission rate is not changed. This is the opportunistic approach, which increases throughput aggressively. 3.2. Expected PU Detection State This state immediately checks the database to verify whether there is any PU activity on the current channel or not. If the database confirms PU activity, it moves to the store state. Otherwise, it goes to the window/rate-based state. 3.3. Store State In this state, the packet loss is due to PU activity and indicates interruption loss. Hence, it keeps the current values of cwnd and the sending rate for further transmission. It is the responsibility of the underlying link layer algorithm to find an alternate vacant channel for transmission or to pause the transmission. 3.4. Window/Rate-Based Transmission State This state is specifically designed to choose either window- or rate-based transmission by looking at cwnd size. According to the empirical results obtained in our simulation, rate-based transmission works well for low cwnd values, and window-based transmission is a better option for higher cwnd values. To achieve the highest throughput and lesser delay, we set 70 as the optimal threshold value from the simulation results presented in the next section. Rate-based transmission is active when cwnd is below 70, and window-based transmission is active at other times. The proposed and compared protocols are designed for interactive streaming applications. It is needless to retransmit the video/voice packet considering the long delay for retransmission. Further, the two common reasons that inhibit the use of TCP for real-time interactive streaming services are Additive-Increase Multiplicative-Decrease (AIMD) and retransmission. In summary, the interactive multimedia applications require the timely packet delivery and smooth predictable transmission rate. 4. Simulation Model and Analysis We conducted an extensive set of simulations in Network Simulator 2 (ns2) [24]. In all the experimental studies, the simulation parameters are the same to provide a fair comparison. As explained earlier, upon the arrival of a PU, the SU has to immediately vacate the channel and look for a vacant channel to resume communications. Otherwise, the SU pauses communications until the PU transmission ends. It is very important to investigate the overall impact of all the transport protocols vigorously in the presence of PU activity. PU activity is independent and random during the simulation. However, it is the same for all the simulations to maintain fairness. The multi-channel CRASN contains 11 channels for communications. All the channels have homogeneous capacity. The parameters used in the simulations are given in Table 1. Table 1. Simulation Parameters. Radio Propagation Model

Two Ray Ground

Channel Type Network Interface Type Mac Type Antenna Model Transport Layer Queue Length Cognitive Radio Model Simulation Time Simulation Area

Wireless Channel Wireless Phy 802.11 Omni Antenna TFRC-CR/TFRC/TFWC/OHTP-CRASN 100 CRAHN 500 s 1000 m ˆ 1000 m

31679

Sensors 2015, 15, 31672–31686

OHTP-CRASN performance depends highly on the threshold value to choose window- or rate-based transmission. We find the optimal value with the help of the empirical results. Figure 8 Sensors 2015, 15, page–page shows time versus throughput for different threshold values. The results indicate that fewer data are delivered when the threshold value is lower. However, it shows improvement with an increasing delivered when the threshold value is lower. However, it shows improvement with an increasing threshold value. Throughput starts to deteriorate after 100. There is not much difference in the results threshold value. Throughput starts to deteriorate after 100. There is not much difference in the results obtained forfor threshold values thelatter latterthreshold threshold value depicts slightly obtained threshold valuesofof100 100and and70. 70. Nevertheless, Nevertheless, the value depicts slightly lessless packet loss and delay. Figure 9 depicts average throughput against the threshold. Therefore, packet loss and delay. Figure 9 depicts average throughput against the threshold. Therefore, according to the simulation thresholdvalue valuetoto according to the simulationoutcome, outcome,we weset set the the optimum optimum threshold 70.70.

Figure8. 8. Optimum Optimum threshold Figure thresholdvalue. value.

Figure averagethroughput. throughput. Figure9.9.Threshold Threshold versus versus average

9 31680

Sensors 2015, 15, 31672–31686 Sensors 2015, 15, page–page

Assume the the scenario scenario of of aa CRASN Assume CRASN in in which which aa set setof ofmultimedia multimediasensors sensorshas hasbeen beenattached attachedwith witha in aa multihop multihop acovert covertbattalion, battalion,which whichtransmits transmitsmultimedia multimediato to the the sink. sink. ItIt generates generates bursty bursty traffic traffic in environment. On a given path, the link with the minimum capacity is called a narrow link, and the the environment. On a given path, the link with the minimum capacity is called a narrow link, and link with the minimum available bandwidth is called a tight link. When the link is both narrow and link with the minimum available bandwidth is called a tight link. When the link is both narrow and tight on on aa given given path, path, itit is is called called aa bottleneck. bottleneck. The The dumbbell dumbbell topology topology [25,26] [25,26] is is the the perfect perfect example example tight of this and is shown in Figure 10. In this topology, m flows are generated at the source nodes and of this and is shown in Figure 10. In this topology, m flows are generated at the source nodes and directed towards sink nodes using the bottleneck link. All flows compete for the finite bandwidth. directed towards sink nodes using the bottleneck link. All flows compete for the finite bandwidth. This realistic realisticnetwork networkscenario scenarioisisdesigned designed evaluate how transport protocols perform under This to to evaluate how thethe transport protocols perform under the the pressure of other sensor nodes with competing traffic that are sharing the same link, and pressure of other sensor nodes with competing traffic that are sharing the same link, and furthermore, furthermore, howisperformance is affected how performance affected by PU activity.by PU activity.

Figure Figure 10. 10. Dumbbell Dumbbell topology. topology.

Figure 11 11 depicts and OHTP. It can be Figure depicts the the instantaneous instantaneousthroughput throughputofofTFRC, TFRC,TFWC, TFWC,TFRC-CR, TFRC-CR, and OHTP. It can seen from the figure that TFRC performs the worst among all the protocols. Even though it is be seen from the figure that TFRC performs the worst among all the protocols. Even though it is designed to to give give aa smoother congestion. It isItevident thatthat the designed smoother transmission transmissionrate, rate,ititfails failstotoidentify identifytrue true congestion. is evident interruption caused by the PU and the resulting packet losses misleads TFRC to detect this as the interruption caused by the PU and the resulting packet losses misleads TFRC to detect this as congestion. As As aa result, result, itit reduces reduces the the transmission transmission rate rate and and eventually eventually underutilizes underutilizes the the bandwidth. bandwidth. congestion. TFWC performs better than TFRC, but it also lacks information to identify PU activity and to adapt adapt TFWC performs better than TFRC, but it also lacks information to identify PU activity and to the transmission rate accordingly. TFWC relies on the loss history. This helps to calculate the average the transmission rate accordingly. TFWC relies on the loss history. This helps to calculate the average loss interval interval (ALI). (ALI). The The calculation calculation of of ALI ALI impacts impacts the the next next cwnd. cwnd. On On the the occurrence occurrence of of packet packet losses, losses, loss it advances cwnd very slowly. Hence, it underutilizes the available capacity. TFRC-CR performs better it advances cwnd very slowly. Hence, it underutilizes the available capacity. TFRC-CR performs than TFRC and TFWC, as can be clearly seen from the diagram. It can differentiate between true better than TFRC and TFWC, as can be clearly seen from the diagram. It can differentiate between congestion and and interruption losses. Therefore, it can handle activity. true congestion interruption losses. Therefore, it can handleand andadapt adaptaccording according to to PU PU activity. However, TFRC-CR reacts very slowly to the immediate decrease in the loss event rate. Subsequently, However, TFRC-CR reacts very slowly to the immediate decrease in the loss event rate. Subsequently, when transmission transmission changes changes to to an an idle idle channel channel with with available available capacity, capacity, itit will will take take aa longer longer time time to to when adjust the rate to the newly available capacity. This is clearly visible from the results presented in adjust the rate to the newly available capacity. This is clearly visible from the results presented in Figure 11. 11. Our the best best among among all all the the competing competing transport transport protocols. protocols. Figure Our proposed proposed OHTP OHTP performs performs the OHTPsmartly smartlydetects detectspacket packetloss lossdue due activity and maintains transmission to exploit OHTP toto PUPU activity and maintains thethe transmission raterate to exploit the the new available capacity. Hence, it achieves greater throughput. Moreover, it adjusts the new available capacity. Hence, it achieves greater throughput. Moreover, it adjusts the transmission transmission rate from true congestion accordingly. This results in an effective utilization of available rate from true congestion accordingly. This results in an effective utilization of available spectrum spectrum andoverall enhances overall system throughput. and enhances system throughput. Table 2 shows the percentage throughput improvement improvement of of OHTP OHTP over over TFRC-CR, TFRC-CR, TFWC, TFWC, and and Table 2 shows the percentage throughput TFRC. OHTP OHTP is is 284.19% 284.19% superior superior to to TFRC. TFRC. Furthermore, Furthermore, OHTP OHTP performs performs 92.51% 92.51% better better than than TFWC. TFWC. TFRC. Our proposed scheme, OHTP, achieves 58.4% higher throughput than TFRC-CR. Hence, the results Our proposed scheme, OHTP, achieves 58.4% higher throughput than TFRC-CR. Hence, the results reaffirm that our proposed OHTP is better than the other compared transport protocols. reaffirm that our proposed OHTP is better than the other compared transport protocols. 10 31681

Sensors 2015, 15, 31672–31686 Sensors 2015, 15, page–page Sensors 2015, 15, page–page

Figure 11. Instantaneous throughput of TFRC, TFWC, TFRC-CR, TFRC-CR, and and OHTP. OHTP. Figure 11. Instantaneous throughput of TFRC, TFWC, TFRC-CR, and OHTP. Table Table 2. 2. OHTP percentage percentage throughput throughput improvement improvement over over transport transport protocols. protocols. Table 2. OHTP percentage throughput improvement over transport protocols.

TFRC TFRC 284.19 284.19

TFWCTFWC TFWC 92.51 284.19 92.51 92.51 TFRC

TFRC-CR 58.4

TFRC-CR TFRC-CR 58.4 58.4

Figure 12 shows the instantaneous delay of OHTP. It is obvious that it incurs greater delay due Figure OHTP. is obvious obvious that incurs greater greater delayatdue due Figure 12 shows instantaneous delay ofitOHTP. ItIt is that itit incurs delay to more waiting time the in the queue. In addition, is expected that high throughput is achieved the more waiting time in the queue. In addition, it is expected that high throughput is achieved at the to time in the queue. In addition, it is expected that high throughput is achieved at cost of high delay [27]. The proposed scheme efficiently maintains the transmission rate under PU cost of high delay [27]. The proposed scheme efficiently maintains the transmission rate under PU the cost of high delay [27]. The proposed scheme efficiently maintains the transmission rate under interruption. Therefore, PU arrival and shifting to the new channel do not have much affect on OHTP. interruption. Therefore, PUreception arrival and shifting to the new do notdo have OHTP. PU Therefore, PU arrival and takes shifting to thechannel new channel notmuch haveaffect muchon affect on As ainterruption. result, the successful of data a longer time. As a result, successful reception of dataof takes longer time. time. OHTP. As a the result, the successful reception dataatakes a longer

Figure 12. OHTP Packet Delay. Figure 12. 12. OHTP OHTP Packet Packet Delay. Delay. Figure

The instantaneous delay of TFRC, TFWC, and TFRC-CR are shown in Figures 13–15, The instantaneous delay of TFRC, TFWC, and TFRC-CR are transmission shown in Figures 13–15, respectively. In TFRC, delay the PU forces the TFRC-CR source to are reduce the rate by half at The instantaneous of activity TFRC, TFWC, and shown in Figures 13–15 respectively. respectively. In TFRC, the PU activity forces the source to reduce the transmission rate by half at every time minimum is reached. As a result, its transmission low time even until after In TFRC, theuntil PU the activity forces rate the source to reduce the transmission rate byrate halfstays at every every time until the minimum rateTherefore, is reached. Asnumber a result, transmission rate stays low even the next available channel is used. the ofits packet transmission is decreased andafter the the next available channel is used. Therefore, the number of packet transmission is decreased and the 11 31682 11

Sensors 2015, 15, 31672–31686

the minimum rate is reached. Sensors 2015, Sensors 2015,15, 15,page–page page–page

As a result, its transmission rate stays low even after the next available channel is used. Therefore, the number of packet transmission is decreased and the delay is reduced. delay reduced. TFWC more throughput than Initially, itit doubles TFWC more delayincurs and higher throughput than TFRC. Initially, it TFRC. doubles the cwnd at each delay isisincurs reduced. TFWC incurs more delay delay and and higher higher throughput than TFRC. Initially, doubles the cwnd at each round trip time. On the occurrence of the first packet loss, it halves the cwnd. round trip time. On the occurrence of the first packet loss, it halves the cwnd. On the occurrence of the cwnd at each round trip time. On the occurrence of the first packet loss, it halves the cwnd.On Onthe the occurrence of the successive packet losses, it advances cwnd very slowly. However, it is still able to the successive packet losses, it advances cwnd very slowly. However, it is still able to achieve better occurrence of the successive packet losses, it advances cwnd very slowly. However, it is still able to achieve throughput TFRC and incurs more TFRC-CR suffers from more delay throughput than TFRC andthan incurs more delay. TFRC-CR suffers from more delay than TFWC the achievebetter better throughput than TFRC and incurs moredelay. delay. TFRC-CR suffers from more delayatthan than TFWC at the cost of more throughput. When the PU is detected, it immediately halts transmission cost of more throughput. When the PU is detected, it immediately halts transmission and restarts TFWC at the cost of more throughput. When the PU is detected, it immediately halts transmission and after PU inactive. Therefore, itit reduces losses, after the PU becomes Therefore, it reduces the packet losses, the and the transmission ratethe is and restarts restarts after the theinactive. PU becomes becomes inactive. Therefore, reduces the packet packet losses, and and the transmission rate isis less affected However, the PU activity queue less affected due PU However, the PU activity causes tocauses build the up the transmission rateto lessinterruption. affected due due to to PU PU interruption. interruption. However, thethe PUqueue activity causes theon queue to up node. Hence, ititattains more delay than node. Hence, attains more delay than TFWC TFRC. to build build upon onitthe the node. Hence, attains moreand delay than TFWC TFWCand and TFRC. TFRC.

Figure Figure 13. TFRC Packet Delay. Figure13. 13.TFRC TFRCPacket PacketDelay. Delay.

Figure Figure14. 14.TFWC TFWCPacket PacketDelay. Delay. Figure 14. TFWC Packet Delay.

31683

12 12

Sensors 2015, 15, 31672–31686 Sensors 2015, 15, page–page Sensors 2015, 15, page–page

Figure15. 15. TFRC-CR Packet Packet Delay. Figure Figure 15.TFRC-CR TFRC-CR PacketDelay. Delay.

Table shows the average delay for TFRC, TFWC, TFRC-CR and OHTP. can clearly be seen Table333shows showsthe theaverage averagedelay delayfor forTFRC, TFRC,TFWC, TFWC,TFRC-CR TFRC-CRand andOHTP. OHTP.ItItItcan canclearly clearlybe beseen seen Table that our proposed scheme, OHTP, incurs the highest delay. However, that the tradeoff between thatour ourproposed proposedscheme, scheme,OHTP, OHTP,incurs incursthe thehighest highestdelay. delay.However, However,that thatisisisthe thetradeoff tradeoffbetween between that high throughput and low packet delay. highthroughput throughputand andlow lowpacket packetdelay. delay. high Table 3. Average Delay (ms). Table Average Delay (ms). Table 3.3.Average Delay (ms).

TFRC TFRC 97.62 97.62

TFWC TFWC 103.70

TFRC 103.70TFWC 97.62

103.70

TFRC-CR TFRC-CR

TFRC-CR 114.87 OHTP 114.87 114.87 175.75

OHTP OHTP 175.75 175.75

InInCRASN, CRASN,nodes nodesdissipate dissipateenergy energyininprocessing, processing,transmitting transmittingand andreceiving receivingmessages. messages.Energy Energy consumption is a key aspect in the design of protocols for sensor networks [28]. This energy needed In CRASN,isnodes dissipate energy in of processing, and receiving messages. consumption a key aspect in the design protocols transmitting for sensor networks [28]. This energyisisEnergy needed for networks. InIn our we the power consumption is a key of aspect in the design of protocols for sensor networks This energy is for correct correct working working of the the sensor sensor networks. our simulation, simulation, we used used[28]. the standard standard power consumption in each state provided in [29]. Figure 16 shows the average energy consumption per needed for correct working the sensor simulation, we used the standard power consumption in each stateof provided in networks. [29]. FigureIn16our shows the average energy consumption per packet ItItis our consumes least consumption inprotocol. each state provided in [29]. Figure 16protocol shows the averagethe energy consumption per packetfor foreach each protocol. isshown shownthat that ourproposed proposed protocol consumes the leastenergy energytototransmit transmit the Hence, ititprolongs the packet for each protocol. It is shown that ourlifetime. proposed thepacket. packet. Hence, prolongs thenetwork network lifetime. protocol consumes the least energy to transmit the packet. Hence, it prolongs the network lifetime.

Figure Figure 16. 16. Energy Energy consumption consumption of of TFRC, TFRC, TFWC, TFWC, TFRC-CR, TFRC-CR, and andOHTP. OHTP. Figure 16. Energy consumption of TFRC, TFWC, TFRC-CR, and OHTP.

31684 13 13

Sensors 2015, 15, 31672–31686

5. Conclusions The introduction of innovative cognitive radio technologies opens up new research areas to overcome the problem of spectrum scarcity by using it more efficiently. We propose the opportunistic hybrid transport protocol (OHTP) for cognitive radio ad hoc sensor networks to achieve the same goal. We designed and embedded the new congestion control algorithm to handle PU arrival intelligently under the transport protocol. Furthermore, there are pros and cons to window- and rate-based transport protocols. For instance, if the network is greatly congested, and as a result the cwnd value remains low, rate-based transmission is well-suited to give smoother throughput. Otherwise, window-based transmission is a better option. Therefore, we propose the hybrid transport protocol. Another issue is to set the optimal threshold value to choose the transmission mode. We empirically calculated and set the threshold value to 70. The results reassure us that we achieved the best outcome. PU interruption detection is also very critical to ensure optimal performance. Helped by this, the transport protocol can differentiate between true and fake congestion. Congestion results in packet losses, and adversely affects the transmission rate of the transport protocol. The phenomenon of packet losses due to PU interruption should not be considered congestion; rather, they are interruption losses. We suggest keeping the same rate, or cwnd, for future transmissions to exploit the next available spectrum. This is the opportunistic way to transmit more packets within the given time. In the case of true congestion, we reduce the rate by the same packet loss ratio. The results indicate that our approach is correct. We compared OHTP with TFRC, TFRC-CR, and TFWC. The results show that OHTP is the superlative transport protocol amongst all of them. Acknowledgments: This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2015R1D1A1A01058751). Moreover, it was supported by the 2015 research grant of Kwangwoon University. Author Contributions: All authors made significant contributions to this paper. Yousaf Bin Zikria proposed the idea, did the simulation and wrote the paper. Summera Nosheen and Farruh Ishmanov provided the valuable insight on transport layer protocols and congestion window mechanism. Sung Won Kim supervised and administered the overall research work. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

Federal Communications Commission. Notice of Proposed Rulemaking and Order (FCC 03-222). Available online: http://web.cs.ucdavis.edu/~liu/289I/Material/FCC-03-322A1.pdf (accessed on 11 December 2015). Mitola, J. Cognitive Radio: An Integrated Agent Architecture for Software Defined Radio. Ph.D. Thesis, KTH Royal Institute of Technology, Stockholm, Sweden, 2000. Akyildiz, I.F.; Lee, W.; Chowdhury, K.R. CRAHNs: Cognitive radio ad hoc networks. Ad Hoc Netw. 2009, 7, 810–836. [CrossRef] Akan, O.B.; Karli, O.B.; Ergul, O. Cognitive radio sensor networks. IEEE Netw. 2009, 23, 34–40. [CrossRef] Joshi, G.P.; Nam, S.Y.; Kim, S.W. Cognitive radio wireless sensor networks: Applications, challenges and research trends. Sensors 2013, 13, 11196–11228. [CrossRef] [PubMed] Postel, J. RFC 793—Transmission Control Protocol. Available online: https://tools.ietf.org/html/rfc793 (accessed on 11 December 2015). Paxson, V.; Allman, M. RFC 2988—Computing TCP’s Retransmission Timer. Available online: https://tools.ietf.org/html/rfc2988 (accessed on 11 December 2015). Fullmer, C.L.; Garcia-Luna-Aceves, J.J. Solutions to hidden terminal problems in wireless networks. ACM SIGCOMM Comput. Commun. Rev. 1997, 27, 39–49. [CrossRef] Jacobson, V. Congestion avoidance and control. ACM SIGCOMM Comput. Commun. Rev. 1995, 25, 157–187. [CrossRef] Floyd, S.; Henderson, B. RFC 2582—The NewReno Modifications to TCP’s Fast Recovery Algorithm. Available online: https://tools.ietf.org/html/rfc6582 (accessed on 11 December 2015). Matins, M.; Floyd, S.; Romanow, A. RFC 2018—TCP Selective Acknowledgment Options. Available online: https://tools.ietf.org/html/rfc2018 (accessed on 11 December 2015).

31685

Sensors 2015, 15, 31672–31686

12. 13. 14. 15. 16.

17. 18. 19. 20.

21.

22. 23.

24. 25.

26.

27. 28. 29.

Brakmo, L.S.; Paterson, L.L. TCP Vegas: End to end congestion avoidance on a global Internet. IEEE J. Sel. Area Commun. 2006, 13, 1465–1480. [CrossRef] Chowdhury, K.R.; Felice, M.D.; Akyildiz, I.F. TCP CRAHN: A transport control protocol for cognitive radio ad hoc networks. IEEE Trans. Mob. Comput. 2013, 12, 790–803. [CrossRef] Tsukamoto, K.; Koba, S.; Oie, Y. Cognitive radio-aware transport protocol for mobile ad hoc networks. IEEE Trans. Mob. Comput. 2015, 14, 288–301. [CrossRef] Salim, S.; Moh, S. A robust and energy-efficient transport protocol for cognitive radio sensor networks. Sensors 2014, 14, 19533–19550. [CrossRef] [PubMed] Floyd, S.; Handley, M.; Padhye, J.; Widmer, J. Equation-based congestion control for unicast applications. In Proceedings of the Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, Stockholm, Sweden, 28 August–1 September 2000; pp. 43–56. Handley, M.; Floyd, S.; Padhye, J.; Widmer, J. RFC 3448—TCP Friendly Rate Control (TFRC): Protocol Specification. Available online: https://www.ietf.org/rfc/rfc3448.txt (accessed on 11 December 2015). Al-Ali, A.K.; Chowdhury, K.R. TFRC-CR: An equation-based transport protocol for cognitive radio networks. Ad Hoc Netw. 2013, 11, 1836–1847. [CrossRef] Jourjon, G.; Lochin, E.; Senac, P. Towards sender-based TFRC. In Proceedings of the IEEE International Conference on Communications( ICC), Glasgow, Scotland, 24–28 June 2007; pp. 1588–1593. Renker, G.; Fairhurst, G. IETF Internet Draft draft-renker-dccp-tfrc-rtt-option-01—Sender RTT Estimate Option for DCCP. Available online: https://tools.ietf.org/html/draft-renker-dccp-tfrc-rtt-option-01 (accessed on 11 December 2015). Choi, S.H.; Handley, M. Designing TCP-Friendly Window-based Congestion Control for Real-time Multimedia Applications. In Proceedings of 7th International Workshop on Protocols for Future, Large-Scale and Diverse Network Transports (PFLDNeT), Tokyo, Japan, 21–22 May 2009; pp. 1–6. Zong, X.X.; Qin, Y.; Li, L. Transport protocols in cognitive radio networks: A survey. KSII Trans. Internet Inf. Syst. 2014, 8, 3711–3730. Federal Communications Commission. Second Memorandum opinion and Order (FCC 10-174). Available online: http://transition.fcc.gov/Daily_Releases/Daily_Business/2013/db0417/FCC-13-53-53A1.pdf (accessed on 11 December 2015). The Network Simulator—NS-2. Available online: http://www.isi.edu/nsnam/ns/ (accessed on 11 December 2015). Hespanha, J.; Bohacek, S.; Obraczka, K.; Lee, J. Hybrid Modeling of TCP Congestion Control. In Proceedings of 4th International Workshop on Hybrid Systems: Computation and Control, Rome, Italy, 28–30 March 2001; pp. 291–304. Huang, Y.; Ghaderi, M.; Towsley, D.; Gong, W. TCP performance in coded wireless mesh networks. In Proceedings of 5th Annual IEEE Communications Society Conference on Sensor, Meshand Ad Hoc Communications and Networks (SECON), San Francisco, CA, USA, 16–20 June 2008; pp. 179–187. Gamal, A.E.; Mammen, J.; Prabhakar, B.; Shah, D. Optimal throughput-delay scaling in wireless networks—Part I: The fluid model. IEEE Trans. Inf. Theory 2006, 52, 2568–2592. [CrossRef] Razzaque, M.A.; Dobson, S. Energy-efficient sensing in wireless sensor networks using compresses sensing. Sensors 2014, 14, 2822–2859. [CrossRef] [PubMed] Feeney, L.M.; Nilson, M. Investigating the energy consumption of wireless network interface in an ad hoc networking environment. In Proceedings of 20th Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM), Anchorage, AK, USA, 22–26 April 2001; pp. 1548–1557. © 2015 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).

31686