Increased Energy Efficiency via Delay-Tolerant Transmissions in Cognitive Radio Networks

Transcription

1 Network Protocol and Algorithm Increaed Energy Efficiency via Delay-Tolerant Tranmiion in Cognitive Radio Network Bi Zhao Centre for Telecommunication Reearch King College London, London, UK Vaili Frideriko Centre for Telecommunication Reearch King College London, London, UK Received: February 15, 2013 Acceted: June 18, 2013 Publihed: June 30, 2013 DOI: /na.v5i URL: htt://dx.doi.org/ / na.v5i Abtract We are currently witneing the emergence of two imortant trend in wirele network, namely the increaed uage of Internet like alication, many of them which are delay tolerant, and cognitive radio technique. In thi aer, the focu i on how to caitalize the delay tolerance of variou alication (uch a , Peer to Peer network, ocial networking & oerating ytem udate and file tranfer to mention jut a few) to reduce the energy conumtion in cognitive network. Since it become feaible to etimate the Primary Uer () connection for the Secondary Uer (SU) by contacting a truted databae containing the information of traffic, a cheme i rooed that exlicitly utilize the ditribution of SU traffic load to rovide load-balancing. By modeling the roblem under invetigation a an M/M/K/L queuing ytem, the erformance of the SU cometition i analyzed under variou traffic blocking threhold and queuing delay. To thi end, by otimizing the frequency channel utilization, the number of SU connection that can be accommodated imultaneouly i derived. A wide et of numerical invetigation reveal how meage tranmiion delay and the ue of available white ace can reduce the energy conumtion without affecting traffic. Keyword: Delay Tolerant Networking, Queuing Theory, Cognitive Radio 31

2 Network Protocol and Algorithm 1. Introduction Sectrum i a fundamental reource in wirele network ince achievable data rate are linearly deendent on the available ectrum in ue. It i widely acknowledged that the utilization of the ectrum can be deemed a are and how ignificant atio-temoral variation. Exloiting thee ectrum white ace, a they are called, ha been the focu of Cognitive Radio (CR). CR network encoma eentially two core functionalitie, namely, ectrum ening and ectrum management [1]. Sectrum ening functionalitie are reonible to detect unued ectrum and a lethora of different method have been conidered to do o which can be broadly taxonomized in cooerative and non-cooerative technique reflecting whether the detection i taking lace by the node itelf or by utilizing other node in the network. Sectrum management on the other hand can be conidered a the deciion engine (mode of oeration and arameter that will be ued for tranmiion) on how to otimize the ue of ectrum baed on the information available from the ectrum ening functionalitie. In eence, CR network oe the otential of imroving ectrum efficiency by allowing the low-riority SU to temorarily eek the wirele ectrum that i licened to different organization () [2]. A oon a the emerge in the frequency channel, the SU mut vacate the licened band. The SU connection would be interruted by the tochatic nature of the traffic. Conequently, the SU hould firtly etimate the channel availability by robability analyi baed on hitorical traffic information or ectrum ening. A common technique for channel etimation ha been to analyze the traffic characteritic from available long-term obervation/tatitic [3]. In [4], author deign otimal ening trategie via a model auming that the tranmiion are unlotted a a continuou-time Markov chain while the SU are lotted to ene the frequency channel. On thi bai, Noh et al. rooe a tochatic multichannel ening cheme baed on traffic information and ening hitory [5]. We aume that the SU can achieve erfect channel etimation about the connection from the Bae Station (BS) or other wirele rovider. In accordance with the information of channel utilization, SU connection can decide which frequency channel are available to utilize without deteriorating the quality of the connection. Queuing theory i a natural mathematical tool to analyze ytem erformance by conidering tochatic arrival and dearture a i the cae in wirele network. A uch, queuing theory ha been alied in ignificant volume of reearch effort within CR literature. In [6], by conidering a two-dimenional M/M/N/N queue, the author have derived the mean number of the and SU with their blocking robabilitie reectively. However, they focued merely on the ituation where the SU connection would be immediately droed if all frequency channel are occuied by in Secondary Uer Cleared (SUC) mode, and the ituation that the and SU connection have equal riority in Secondary Uer Equality (SUE) mode. In our model, we aume that there are M/M/K/L queue ytem for the SU connection, in which the SU meage could be buffered. A reemtive riority queuing ytem ha been utilized to analyze the mean ytem dwelling time of the SU traffic and the blocking robability for real-time SU connection [7]. In [8], the author analyze the queue length and average queuing delay of the SU baed on Poion ditribution of the SU. In [9], 32

3 Network Protocol and Algorithm a Dynamic Strategy Learning (DSL) algorithm relied on the riority queuing ytem including the SU and the i rooed for the delay-enitive multimedia alication in order to maximize the uer utility function. Conidering that an increaing number of countrie are now ermitting oeration of cognitive radio ytem in the vacant terretrial analog TV tranmiion - TV White Sace (TVWS), it will create new oibilitie to rovide better wirele broadband and multimedia ervice [10]. When the SU are uing 3G cellular network within the coverage of a TVWS mater, it i oible that the SU would refer the TVWS connection over the cellular network in term of cot, RF coverage, caabilitie, and overall Quality of Service. The SU connection can utilize the mean a ening or contacting a truted geoatial databae that record the information regarding occuation with a ecific location and time duration, rior to meage tranmiion, to determine available ectrum at a given location [11]. Moreover, when the SU are vehicular uer, they may want to build the connectivity to another wirele node in eer-to-eer or ad-hoc network. In thi cenario, to redict the future location and the ath of mobility of wirele node i another challenging iue in the o-called White-Fi network. In our context, in order to caitalize the delay tolerance of alication, we develo a theoretical framework for energy efficiency cheme that can maximize ectrum utilization and SU throughut. Our aroach relie on Adative Modulation and Coding (AMC) to dynamically change the modulation and coding cheme. Once the mobile device are equied with a dual or multi mode antenna that i connected to 3G and other network like TVWS, the SU could witch between thee network to eek and ue any licened ectrum band a long a they do not caue interference to the. By conidering the ditribution of the SU traffic load and connection that would be emerging tochatically, the CR ytem will contact a truted databae for hitorical information about traffic at a ecific location and time duration o a to etimate the robability for the SU connection. Baed on the etimation of traffic within a ecialized relatively long-term duration, and the analyi of robabilitie for vacant channel and time lot, an M/M/K/L queuing ytem i devied to etimate the SU traffic caability that the ytem can erve imultaneouly. By conidering an M/M/K/L queue where SU meage comete for K frequency channel in N concentric ring with different modulation and coding cheme within the BS coverage, the traffic of the SU can be inutted to the queue for the SU connection. If the number of SU i large, the inut traffic of the virtual queue can be modeled a a Poion roce, where L i the finite number of waiting oition for each queue. Finally, for the delay-tolerant alication, we recommend the ytem to delay the meage tranmiion to an area cloe to the BS, thereby maximizing the throughut otential whilt reducing the overall energy cot of meage tranmiion under everal contraint. The ret of the aer i organized a follow. In Section II, the related work on ectrum election cheme and queuing model in CR are ummarized. Section III define the model of energy conumtion over cellular network, and formulate the otimal SU connection aroximation in the queuing ytem of CR network. Section IV reent numerical invetigation and analyi of otimal tranmiion cheme. Finally, we conclude the work 33

4 Network Protocol and Algorithm in Section V. 2. Related Work 2.1 Delay Tolerant Network Significant volume of reearch effort have been laced on energy efficient data tranmiion for delay tolerant alication, eecially looking at the trade-off between tranmiion cot and delay tolerance over wirele network [12] [13] [14] [15]. The work in [12] deal with the roblem of acket cheduling with deadline within a re-defined time window [0, T). Baed on that, the author in [16] exlore the energy-efficient acket tranmiion with individual acket delay contraint, in which a trade-off between flexible energy and delay i analyzed under variou individual acket delay contraint and bandwidth efficiencie. In [13], the author conider a delay contraint for each acket and reveal the relationhi between reliable tranmiion rate and QoS requirement, while a dynamic rogramming baed algorithm i introduced to acquire throughut maximization and energy minimization according to different channel qualitie of a fading channel with time contraint [14]. The work in [15] further invetigate the roblem of energy-delay trade-off under dynamic traffic load and uer oulation. The target-et election roblem ha been tudied in the emerging Mobile Social Network for traffic offloading by delaying the delivery [17]. In [18], a framework i rooed to invetigate the trade-off between the amount of offloaded traffic and the uer delay tolerance over a 3G network. 2.2 Channel Selection There have been everal rior work on dynamic ectrum acce, eecially channel election cheme. In [19], a dynamic rotocol i introduced for dynamic ectrum allocation (DSA) with load balancing for SU. They aume that the traffic detection and SU connection blocking in thee channel from the allocation game ha been olved already, therefore they jut focu on the load balancing of SU traffic. Their algorithm are modeled a a o-called ball and bin roblem of congetion game, where ball rereenting the CR are aigned to everal bin rereenting hyical channel of the radio ectrum and SU may reaign it load in a round baed fahion dynamically. The ytem can raidly converge toward an aroximately balanced tate in which all CR will utain cot below a certain threhold arameter. In [20], author rooe an effective ectrum deciion cheme, which can evenly ditribute the traffic load of SU connection to multile channel, thereby reducing the average overall ytem time comared to the non-loadbalancing cheme. They dicu two kind of ectrum deciion cheme: one i ening-baed with the objective to determine the otimal ened number of candidate channel for channel election, the other i robability-baed that take the traffic tatitic of both and SU. In [21], the author deign a minimum colliion rate algorithm and minimum handoff rate algorithm to maximize ectrum hole utilization of the channel for otimal SU throughut on the bai of 34

5 Network Protocol and Algorithm atifying contraint of colliion tolerable level of channel. Thee two channel election cheme are baed on ectrum hole rediction from at obervation. In [22], a oortunitic channel election cheme i rooed with the aid of tatitical traffic attern of channel and traffic rediction technique. According to the long-term tatitical robability of each channel aearing idle in the next time lot, SU will firtly ene the channel with the highet average robability of being idle. Likewie, a channel election method ha been devied in [23] to analyze the mot robable unoccuied channel for SU traffic. When a databae receive a SU query for an unoccuied channel, it will deliver tored hitorical information about channel availability. The databae earche the channel that have been unoccuied at reviou day from that time lot to ome time to future, while it enure noie level of the mot robably vacant channel i lower than a threhold. In [24], in order to achieve better channel utilization, SU need to make a deciion at each time lot to acce if the current achievable throughut can be imroved by utilizing another hyical channel baed on channel ening and etimation. Thi channel exloitation roblem can be modeled a a finite-horizon otimal toing roblem between the exected increaing data rate of SU and channel ening cot. 2.3 Queuing Theory in CR Many different facet of queuing theory have been utilized in the CR literature to analyze the different rioritie cheme for and SU traffic. In [25], a lotted tranmiion ytem and an infinite queue are aumed for both and SU traffic. SU ene the channel in each lot, and tranmit a acket from the correonding queue if detecting an idle time lot. In thi context, SU link ha been conidered a a tranarent relay for the traffic, if it doe not affect the tability of the queue. In [26], the reemtive reume riority (PRP) M/G/1 queuing network model i develoed to characterize the channel uage of CR network. Baed on thi queuing model, they characterize the ectrum utilization behavior of each channel with different arrival rate and ervice time ditribution of and SU. The reearch invetigate the effect of ectrum hand-off delay on the extended data delivery time of the SU connection. A ignificant volume of reearch took the aroach of categorizing uer into K riority clae. In that cae, the highet riority cla C 1 i alway reerved for the in each channel and SU occuy the ret of K 1 riority clae ( C2, C3,, Ck ) [9] [27]. In [9], an M/G/1 model i adoted and each SU maintain everal hyical (or logical) queue for the variou frequency channel. The channel election deciion are baed on riority queuing analyi that conider the dynamic channel condition, traffic characteritic, and the cometitor behavior. Hence, the SU could efficiently adat their channel election trategie. In [27], they alo model the traffic a an M/G/1 reemtive riority queue with channel condition and one and multile SU cometing for the ame frequency channel. Vacating the channel for the traffic, SU would ene the remained channel and time lot (ectrum hole) according to a imle Firt-Come-Firt-Serve (FCFS) rule. Moreover, they lace the emhai on the delay and throughut of the SU traffic intead of tability. An M/M/1 time-varying queuing model i introduced to generate an accurate temoral 35

6 Network Protocol and Algorithm and frequency characterization of CR network in [28]. In thi cae, the ingle erver i aumed to be a centralized BS and ectrum occuancy and availability for SU connection i baed on a tatitical model. In [29], the author deign a model that occuy many licened channel with one channel allocated for SU traffic. The SU traffic ubytem i modeled a an M/M/1 queue with a roceor haring (PS) olicy. Meanwhile, the SU would ene the licened channel in order to find a free channel for tranmiion. A a reult, the traffic of licene channel could be modeled uing an M/M/K/K queue. traffic in three different level of riority comared to SU connection (erfect, artial and no riority) ha been analyzed in [6], where an M/M/N/N queuing ytem i utilized to analyze and SU traffic of arrival roce and ervice rate. In [7], the reemtive riority queuing model i deigned to analyze the SU traffic of CR ytem. The ditributed non-real-time SU traffic enter a queue buffer in Firt-In Firt-Out (FIFO) manner when the erver i buy with or earlier SU connection. In non-real-time ituation, SU traffic are allowed to be buffered in queue. Meanwhile, if there i no erver available, the real-time SU traffic are immediately dicarded. They how the imulation reult regarding ytem dwelling time for non-real-time SU connection and blocking robability and forced termination robability for real-time SU traffic under different SU arrival rate. 3. Sytem Model Hereafter, a ingle cell wirele network i conidered where Adative Modulation and Coding (AMC) are uorted. To thi end we denote with { M R, M,, } 1 R M 2 R the et of N available modulation and coding cheme which can be elected in eence according to the ditance between the wirele device and the BS. In that reect, the cell can be enviioned a being earated into n concentric ring of radii R i, i 1,2,, N a hown in Fig. 1. In thi cae, each circular region with ditance R i to the BS correond to a different contellation ize and coding cheme. SU 64QAM-3/4 64QAM-2/3 16QAM-3/4 16QAM-1/2 QPSK-3/4 QPSK-1/2 Figure 1. Grahical rereentation of a cell decomoed into ix concentric ring correonding to different tranmiion rate 36

7 Network Protocol and Algorithm Let r denote the coding rate, and r BS be the (Euclidean) ditance between the mobile node and the erving BS. The ectral efficiency (bit//hz) i given by [30]: IEC( rbs ) r log 2( M r ) bit / / Hz BS (1) Concerning the energy conumtion in meage tranmiion, we identify the following main ource: (1) the electronic circuit at the node of SU; (2) the energy diiated to tranmit the meage; and (3) the energy conumed to receive correonding meage. The energy conumtion regarding SU wirele meage tranmiion can be formulated a in [31], hown in the equation below where e d rereent the energy conumtion of the o-am and B rereent the data tranmiion rate. f ( r ) ( e e e r B (2) BS rx tx d BS ) ( e ( e rx rx e e tx tx e e lo m r r 2 BS 4 BS ) B ) B if if r r BS BS r r th th (3) where e lo denote the one bit energy conumtion with the free ace (line-of-ight) roagation loe (the ditance to the BS rbs rth) and e m i the energy conumed er bit in tranmiion in the longer-ditance cae that the ignal i attenuated by multi-ath fading (Two-ray Reflection). Let P lo and P m denote the tranmitted ignal ower in thee two cae reectively, we obtain P lo P (4 ) G G rx 2 wl tx 2 rx r 2 BS (4) P m Prx 4 r 2 BS (5) ( h h ) G G tx rx tx rx where wl i the wavelength, h tx / h rx i the tranmit/receive antenna height, and G tx / Grxi the tranmitter/receiver antenna gain. Therefore, the energy conumtion for tranmiion i given by: k m t k f ( r, t) ( e e ) F P ( or P ) t (6) BS rx tx lo m where km i the time lot that finih the meage tranmiion Let { k, k1, k2,, km} denote the time unit for meage tranmiion in one circular region, where k i the time unit to launch the meage tranmiion, and km i the time unit that comlete the tranmiion. In each time unit, one meage block will be 37

8 Network Protocol and Algorithm tranmitted to the erving BS. Conequently, the length of time unit in each ring hould be different due to the different throughut achieved in each of the different circular area of the cell. Within cognitive radio (CR) network, the SU tranmiion i trying to exlore white ace of wirele ectrum while at the ame time avoiding interfering with the traffic. For the uroe of avoiding tranmiion, a a SU, the fundamental aumtion i that wirele node in vehicle ue ectrum ening or query the databae which maintain information about the available channel for the detail of the local radio environment. 3.1 Otimal Tranmiion Time of SU In the rooed model a detailed hereafter we aume that wirele node gather traffic information from a hitorical databae in order to redict over a hort-term the traffic attern of. In the databae erver, there are two tye of information about rimary channel. One i the 24-hour traffic characteritic of different channel in everal month [3], the other i the noie ower level in different channel that udated by SU device contantly. Baed on the above framework, a SU ha to end a query to databae erver for the available channel to tranmit. According to the available channel information at the ame time lot in reviou day and long-term tatitic regarding channel availabilitie, the databae erver will certificate the noie ower level by comaring to a threhold. Then the bet candidate channel for the inquired SU will be determined. A there i ignificant reviou reearch regarding the rediction of connection arrival rate and holding time, we aim to utilize uch reult and deign an algorithm for SU connection according to the eriodicity of traffic attern. In thi cenario, we conider the time horizon for the etimation to be relatively long term (uch a for examle in term of hour) intead of hort term (fraction of a econd), that i, the rediction of traffic i table and unchangeable in a relative long-term time. Therefore, under the aumtion of the robability-baed channel election cheme [20], the SU will elect it oerating channel from all of the M candidate channel to achieve ome form of load-balancing baed on the tochatic traffic rediction, uch a the arrival rate and ervice time. Alying the Probability-baed Scheme theory [20], the average ytem time S i given by, E[ S] E[ W] E[ T] (7) where W denote the waiting (queuing) time and T reent the extended data tranmiion (k ) (k ) time. Then, let X and X be the extended ervice time of and SU reectively due (k ) to the mied detection and fale alarm robabilitie. and are the average arrival (k ) (k ) rate of the connection at channel k and the SU connection. and are the buy robabilitie reulting from the and SU connection at channel k. E[ T] K k1 ( k ) b ~ ( k ) E[ X ] ( k ) 1 (8) 38

9 Network Protocol and Algorithm E[ W ] K k1 ( k ) b 1 2 ~ E[( X ( k ) ( k ) 2 k ) ( (1 ) ] 1 2 ( )(1 ( k ) k ) ~ E[( X ( k ) ) ) ( k ) 2 ] (9) ( k) ( k) ~ ( k) E[ X ] (10) ( k) ( k) ~ ( k) E[ X ] (11) ( 1) (2) ( k) ( K ) where the vector b (,,,, ) reent the et of ditribution robabilitie of candidate channel. We utilize a robability-baed cheme to find the otimal robability vector b in order to balance the and SU traffic load among multile (1) (2) ( K ) channel with the contraint that 1. Moreover, thi otimal b vector can maximize the tranmiion time in each time lot in order to accommodate more SU traffic. On the bai of thi otimal vector, the SU could directly acce the reelected channel by etimating the traffic characteritic from hitorical tored data. From Fig. 2, we can oberve that when the SU traffic i relatively light, i.e., the arrival rate being 0.01, the SU traffic refer to elect channel 1 a their tranmiion frequency channel which ha the lightet traffic. However, when the SU arrival rate become higher a the arrival rate i 1, all of the channel eem to be identical for the SU traffic, and the SU connection don t have any riority over the candidate channel for meage tranmiion. In our model, it i aumed that the average arrival rate of the SU connection are under a threhold which retrict the SU traffic for better QoS ervice Otimal Ditribution Probability Vector[] Channel 1 Channel 2 Channel 3 Channel 4 Time duration in one time lot Otimal Rate In time lot for Tx Average Arrival Rate of the SU Connection Figure 2. SU traffic attern with different average arrival rate Recall that the cell ha been earated into N concentric ring a hown grahically in Fig. 1. When a SU meage need to be tranmitted, firtly, it will be laced into a ecialized queue for correonding ring. Secondly, the ytem of queue i ening for the available 39

10 Network Protocol and Algorithm vacant frequency channel that which the connection do not occuy and trie to ditribute the meage from SU queue into the candidate channel. We aume without lo of generality that all SU meage have the ame data length. A a reult, the SU connection in all of the frequency channel are aumed to have the ame ervice time. From the above dicuion, it i oible to etimate the otimal time duration for SU traffic in a ecific time lot. Then the roblem can be imlified into a ituation that only SU meage comete for all of the frequency channel, i.e., without cometing with the connection. Under the above defined framework, the central focu in the equel i to calculate the number of SU that can be acceted in the different co-eccentric ring in the network under the contraint of ervice delay elaticity and the blocking robability. 3.2 Otimal Number of SU In thi ection, we turn our focu on the SU meage cometition without cometing connection. Fig. 3 give an examle of the hyical queue for the cae of K frequency channel and N concentric ring with different modulation and coding cheme. When the traffic of the SU need to be tranmitted in the ytem, it can be inutted to the queue for the SU connection. Thi rooed channel election model could aroximate the virtual SU meage queue uing an M/M/K/L queuing ytem. If the number of SU i large, the inut traffic of the virtual queue can be modeled a a Poion roce, where K i the number of erver and L i the finite number of waiting oition for each queue. (1) M/M/K/L Queuing Sytem C11 C12 Queue with L lace C1K (2) C21 C22 L C2K (N) CN1 CN2 L CNK Figure 3. Acce for SU modelled an M/M/K/L queuing ytem (i) Let denote the average number of the SU er unit time in ith concentric ring of radii R i and L denote the number of unit time a a kind of queue length. Therefore, each ( queue of thi ytem, namely each ring, can accommodate i ) L number of the SU. Given the et of candidate channel { 1,2,, K} and the et of concentric ring { 1,2,, N}, we denote C ij to be the caability of the SU in ring i within the channel j and have 40

11 Network Protocol and Algorithm C ij B IEC( rbs ) (bit/) (12) K N where F i the ize of SU meage and B rereent the bandwidth available at the BS. Let ij rereent the ervice rate of SU connection uing the frequency channel j in ith ring, we have, C ij ij (13) F Let i denote the occuation rate (offered traffic load), we have ( i) ( i) F i (14) K K C ij ij Aume that m i the robability that there are m SU meage in the ytem and the blocked traffic rate threhold, therefore we have thre i m m i 0 m! m i i ( ) K! K mk 0 K m K ( i) m L K (15) ubject to: ( i ) L K m m0 1 (16) i ( ) ( L K ) thre (17) Note that the contraint of (17) deict that the blocking rate of SU connection in the virtual queue hould be lower than the redetermined threhold. thre 4. Numerical Invetigation Table 1.Sytem Parameter. Symbol Definition Value P tx Received ower threhold -52 dbm va Vehicle average eed 8 m/ BW Bandwidth 10 MHz e tx Tranmitter electronic conumtion J/bit e rx Receiver electronic conumtion J/bit 41

12 Network Protocol and Algorithm The related arameter ued in general cenario where a vehicle i moving toward the BS are ummarized in Table 1 and we ue the model in [32] to calculate the energy conumtion of RF module and electronic circuit. The reult of energy conumtion were obtained via MATLAB baed imulation by focuing on the ulink throughut to erving BS and it i aumed that the radiu of the cellular BS to be 1000 meter. We rovide an examle of otimal allocation of the connection and the blocking rate robabilitie of SU connection in Table 2. For imlicity, we aume the channel number K = 4. Note that in Table 2, the arrival rate and ervice time of the connection in all of the K channel are not identical. In addition, a hown in the reult of otimal ditribution robability vector for SU tranmiion, we can ee that increaed robabilitie of SU connection are allocated to the channel whoe traffic load i exected to be le. A a reult, the wirele node on the vehicle would alway try to elect the otimal channel for data tranmiion and avoid the interrution on the tranmiion. Baed on the above dicuion, the otimal ditribution robability vector for tranmiion time of vehicle ot i (0.2968, , , ). Table 2.Parameter of Traffic. Num of Channel Arrival rate ( ) Service time ( X ] E [ ) Probability of SU traffic (P ot ) Recall that the SU mut monitor all frequency channel to ene the arrival of connection. In thi model, we aume that the traffic load i table within a hort time duration. Therefore, from now on, we ue queuing ytem modeling to analyze the traffic of SU connection without conideration of connection. The SU virtual queue ene the frequency channel in an increaing order, from 1th to the Mth channel. When initiating a meage tranmiion, the queue firt check the availability of frequency channel 1. If all of the frequency channel are unavailable, the tranmiion i blocked. Here, for the SU virtual queue, the availability of one frequency channel mean that the channel i not occuied by the connection. Table 3 reent the imulation arameter and reult with a blocking traffic threhold thre = 0.10 for the SU queue. It can be clearly een from the reult that when the mobile node are moving toward the ervice BS, the length of time unit in each ring become le owing to the increaing caability of ring in term of tranmiion rate. Moreover, with the decreaing ditance between wirele node and the BS, the ring could accommodate a larger number of SU meage, that i, the SU virtual queue could accommodate a larger number of SU imultaneouly. 42

13 Network Protocol and Algorithm Table 3.Simulation Reult ( thre =0.10). Num of Ring Num of time unit Max Packet length F (Mbit) Blocking Probability total N/A otimal In thi model, we aume that there are two ituation for a comarion in order to how the benefit of energy conumtion: 1) the bandwidth of the ervice BS i divided into ix equal art, and each ring could only ue one art for tranmiion; 2) the bandwidth i occuied excluively by the lat ring which i cloet to the ervice BS. For intance, in the 7th line of table 3, the SU buffer all the meage and move into the area of the ring which i cloet to the ervice BS. In thi cae, the SU in the lat ring occuy all of the bandwidth for meage tranmiion, that the caacity and throughut will be the maximum oible ince thi ring can uort the higher contellation. From Fig. 4 we can oberve that, if the meage carried by the SU i highly delay-tolerant, when the SU tranmit meage in the cloet ring of the BS, the throughut could be ignificantly imroved. Mean Throughut[kt/ ec] tranmiion inall of thering tranmiion in thecloet ring BlockingProbability Figure 4. Mean throughut of the SU v. different blocking robability threhold Concerning the energy conumtion, Fig. 5 reent three different cae of energy cot focuing on one SU to be erved. Firtly, when the wirele node i moving toward the BS, we comute the energy conumtion in all the ix ring of the BS. A econd cae hown i the 43

14 Network Protocol and Algorithm ituation when the SU buffer the meage to be tranmitted to the BS and move into the lat ring for data tranmiion, which the SU occuie all of the bandwidth of the BS. Latly, and baed on the econd cae, the rooed cheme i deicted where the SU ha moved into the lat ring of the BS, a the bandwidth increaing, the SU queuing ytem can accommodate an increaed number of SU meage under a threhold of blocking traffic rate, which bring all the caabilitie into full lay. Here, we mut emhaize that in the ituation of the firt and third cae, the queuing ytem work under an identical blocking robability threhold to retrict the meage number in the queue. A a reult, intuitively eaking, the otimal one (the third cae) ha the otential of increaing the energy efficiency due to the fact that it can roce a higher number of SU meage imultaneouly; thu ending le time to tranmit the meage comaring to the ub-otimal one (conidering meage of the ame ize). From Fig. 5, it i hown that, a the threhold of blocking traffic robability i et to be 10%, the otimal cheme only diiate aroximately 2/5 energy (Joule) with equal amount of meage tranmiion. It i noteworthy that, the energy cot of otimal cheme fluctuate narrowly among thee three cheme under different blocking robability threhold. The reaon i that the otimal cheme rovide the highet data throughut and accommodate larget number of SU meage, which lead to the hortet time duration for wirele tranmiion and circuit cot. Figure 5. Energy conumtion of tranmiion in different blocking robability threhold for one SU It i aumed that the length of each meage i identical. Due to the different throughut of wirele tranmiion in different ring of area, the length of time unit that the wirele node could end one meage to BS hould be different a hown in Fig. 6. Meanwhile, with the monotonically increaing value of blocking robability threhold, the number of SU meage that can be accommodated in the queuing ytem will rie in varying degree. Comared to the firt two cae, the otimal cheme can accommodate ignificantly more meage in the queuing ytem imultaneouly a hown in Fig

15 Network Protocol and Algorithm Figure 6. The length of time unit in different ring Figure 7. The max value of _ in different ring under different blocking robability threhold Additionally, the cenario invetigated above can be extended for the cae where there are multile SU tranmitting meage imultaneouly. For illutration uroe, we et a blocking traffic threhold thre = 0.10 for the SU queue in thi imulation a well. A reented in Fig. 8, the trend of three curve are almot the ame a before and the rooed cheme outerform the other cheme. 45

16 Network Protocol and Algorithm Figure 8. Energy conumtion of tranmiion with increaing SU number 4. Final Remark and Concluion In thi aer, a cheme i rooed for delay-tolerant Internet alication in CR network with the central aim of reducing the energy conumtion; and a a reult to rolong the recharging eriod of terminal. Conidering the tochatic ditribution of the arrival rate and time duration in rimary channel, the SU ha to query a truted databae for hitorical information in order to elect candidate channel for meage tranmiion. With the aid of the etimated available robability concerning location and time duration, the SU traffic can acce the rimary channel more efficiently and increae the channel utilization. Furthermore, we took the M/M/K/L queue to derive the average number of imultaneou SU meage and correonding blocking robabilitie for different ituation. When non-real-time alication are buffered in the SU ytem, our otimal cheme have an inclination to delay the meage tranmiion to the area with higher throughut, i.e., cloer to the BS. Our analyi and numerical reult reveal that the rooed otimal meage tranmiion cheme can accommodate ignificantly more meage in the queue imultaneouly under the ame blocking threhold. Alo, due to the highet data throughut of the rooed cheme, the overall energy cot of meage tranmiion can alo be ignificantly decreaed. Finally, when the cenario i extended to multile SU cae, the rooed cheme till outerform the other cheme. In the future, once the detail of databae regarding traffic i available for ractical ue, we lan to tudy in a more detailed manner the technique that ue the available ga in radio ectrum, (the o-called white ace ), which exit in the band that have been reerved for analog TV broadcating. 46

17 Network Protocol and Algorithm Reference [1] Simon Haykin, Cognitive Radio: Brain-Emowered Wirele Communication, IEEE Journal on Selected Area in Communication, vol.23, no. 2, , February htt://dx.doi.org/ /jsac [2] Ian F. Akyildiz, Won-Yeol Lee, Mehmet C. Vuran, and Shantidev Mohanty, NeXt generation/dynamic ectrum acce/cognitive radio wirele network: a urvey, Comuter Network: The International Journal of Comuter and Telecommunication Networking, vol. 50, no. 13, , Set htt://dx.doi.org/ /j.comnet [3] Xiukui Li and Seyed A. Zekavat, Traffic Pattern Prediction and Performance Invetigation for Cognitive Radio Sytem, Wirele Communication and Networking Conference (WCNC), , Mar htt://dx.doi.org/ /wcnc [4] Qianchuan Zhao, Stefan Geirhofer, Lang Tong and Brian M. Sadler, Oortunitic Sectrum Acce via Periodic Channel Sening, IEEE Tranaction on Signal Proceing, vol. 56, no , Feb htt://dx.doi.org/ /tsp [5] Goan Noh, Jemin Lee and Daeik Hong, Stochatic Multichannel Sening for Cognitive Radio Sytem: Otimal Channel Selection for Sening with Interference Contraint, Vehicular Technology Conference Fall (VTC 2009-Fall),. 1-5, Set htt://dx.doi.org/ /vetecf [6] Peter J. Smith, Abdulla Firag, Pawel A. Dmochowki and Manoor Shafi, Analyi of the M/M/N/N Queue with Two Tye of Arrival Proce: Alication to Future Mobile Radio Sytem, Journal of Alied Mathematic, htt://dx.doi.org/ /2012/ [7] Junghyun Heo, and Jungchae Shin et al., Mathematical Analyi of Secondary Uer Traffic in Cognitive Radio Sytem, Vehicular Technology Conference 2008 (VTC 2008-Fall),. 1-5, Set htt://dx.doi.org/ /vetecf [8] Shanhan Wang, Junhan Zhang and Lang Tong, A Characterization of Delay Performance of Cognitive Medium Acce, IEEE Tranaction on Wirele Communication, vol. 11, no. 2, , Feb htt://dx.doi.org/ /twc [9] Hien-Po Shiang and Mihaela van der Schaar, Queuing-Baed Dynamic Channel Selection for Heterogeneou Multimedia Alication Over Cognitive Radio Network, IEEE Tranaction on Multimedia, vol. 10, no. 5, , Aug htt://dx.doi.org/ /tmm [10] Stehen J. Shellhammer, Ahmed K. Sadek and Wenyi Zhang, Technical challenge for cognitive radio in the TV white ace ectrum, Information Theory and Alication Workho, , Feb htt://dx.doi.org/ /ita [11] S. Probaco, Ed. B. Patil, Protocol to Acce White Sace databae: PS, ue cae and rqmt, online at: htt://tool.ietf.org/html/draftietf-aw-roblem-tmt-uecae-rqmt-03. [12] E. Uyal-Biyikoglu, B. Prabhakar, and A. El Gamal, Energy-efficient acket tranmiion over a wirele link, IEEE/ACM Tranaction on Networking, vol. 10, no. 4, 47

18 , htt://dx.doi.org/ /tnet Network Protocol and Algorithm [13] Xiliang Zhong, and Cheng-Zhong Xu, Delay-Contrained Energy-Efficient Wirele Packet Scheduling with QoS Guarantee, GLOBECOM 2005, vol. 6, , htt://dx.doi.org/ /glocom [14] Alvin Fu, Eytan Modiano, and John Titikli, Otimal Energy Allocation for Delay-Contrained Data Tranmiion over a Time-Varying Channel, INFOCOM 2003, vol. 2, , July htt://dx.doi.org/ /infcom [15] Hongeok Kim, and Gutavo de Veciana, Leveraging dynamic are caacity in wirele ytem to conerve mobile terminal energy, IEEE/ACM Tranaction on Networking (TON), vol. 18, no. 3, , June htt://dx.doi.org/ /tnet [16] Wanhi Chen, Michael J. Neely, and Urbahi Mitra, Energy-Efficient Tranmiion With Individual Packet Delay Contraint, IEEE Tranaction on Information Theory, vol. 54, no. 5, , May htt://dx.doi.org/ /tit [17] Bo Han, et al., Cellular traffic offloading through oortunitic communication: a cae tudy, Proceeding of the 5th ACM workho on Challenged network (CHANTS 10), New York, NY, USA, htt://dx.doi.org/ / [18] Xuejun Zhuo, Wei Gao, Guohong Cao, and Yiqi Dai, Win-Couon: An incentive framework for 3G traffic offloading, Network Protocol (ICNP), , Vancouver, BC, Oct htt://dx.doi.org/ /icnp [19] Simon Ficher, Petri Mahonen, Marcel Schongen, and Berthold Vocking, Load Balancing for Dynamic Sectrum Aignment with Local Information for Secondary Uer, New Frontier in Dynamic Sectrum Acce Network (DySPAN 2008), Chicago, USA,. 1-9, Oct htt://dx.doi.org/ /dyspan [20] Li-Chun Wang, et al., Load-Balancing Sectrum Deciion for Cognitive Radio Network, IEEE Journal on Selected Area in Communication, vol. 29, no. 4, , Aril htt://dx.doi.org/ /jsac [21] Qinghai Xiao, Yunzhou Li, Ming Zhao, Shidong Zhou, and Jing Wang, Oortunitic channel election aroach under colliion robability contraint in cognitive radio ytem, Comuter Communication, vol. 32, no. 18, , Dec htt://dx.doi.org/ /j.comcom [22] Guangxiang Yuan, Ryan C. Grammeno, Yang Yang, and Wenbo Wang, Performance Analyi of Selective Oortunitic Sectrum Acce With Traffic Prediction, IEEE Tranaction on Vehicular Technology, vol. 59, no. 4, , MAY htt://dx.doi.org/ /tvt [23] Johanna Vartiainen, Marko Hoyhtya, Janne Lehtomaki and Timo Brayy, Priority Channel Selection Baed on Detection Hitory Databae, 2010 Proceeding of the Fifth International Conference on Cognitive Radio Oriented Wirele Network & Communication (CROWNCOM), Canne,. 1-5, June

19 Network Protocol and Algorithm [24] Ho Ting Cheng, and Weihua Zhuang, Simle Channel Sening Order in Cognitive Radio Network, IEEE Journal on Selected Area in Communication, vol. 29, no. 4, , APRIL htt://dx.doi.org/ /jsac [25] Ovaldo Simeone, Yehekel Bar-Ne, and Umberto Sagnolini, Stable Throughut of Cognitive Radio With and Without Relaying Caability, IEEE Tranaction on Communication, vol. 55, no. 12, , Dec htt://dx.doi.org/ /tcomm [26] Chung-Wei Wang, and Li-Chun Wang, Analyi of Reactive Sectrum Handoff in Cognitive Radio Network, IEEE Journal on Selected Area in Communication, vol. 30, no. 10, , NOVEMBER htt://dx.doi.org/ /jsac [27] Caoxie Zhang, Xinbing Wang, and Jun Li, Cooerative Cognitive Radio with Priority Queueing Analyi, IEEE International Conference on Communication (ICC 09), Dreden,. 1-5, June htt://dx.doi.org/ /icc [28] Chittabrata Ghoh, Srikanth Pagadarai, Dharma P. Agrawal, and Alexander M. Wyglinki, Queueing Theory Rereentation and Modeling of Sectrum Occuancy Emloying Radio Frequency Meaurement, 2009 IEEE 70th Vehicular Technology Conference Fall (VTC 2009 Fall), Anchorage, AK,. 1-5, Set htt://dx.doi.org/ /vetecf [29] Ouama Habachi, and Yezekael Hayel, Otimal Sening Strategy for Oortunitic Secondary Uer In a Cognitive Radio Network, Proceeding of the 13th ACM international conference on Modeling, analyi, and imulation of wirele and mobile ytem (MSWIM 10), Bodrum, Turkey, , Oct htt://dx.doi.org/ / [30] Rainer Schoenen and Bernhard H. Walke, On PHY and MAC erformance of 3G-LTE in a multi-ho cellular environment, Wirele Communication, Networking and Mobile Comuting (WiCom) 2007, Shanghai, China, , Set htt://dx.doi.org/ /WICOM [31] Bi Zhao, and Vaili Frideriko, Otimal Stoing for Energy Efficiency with Delay Contraint in Cognitive Radio Network, IEEE 23rd International Symoium on Peronal, Indoor and Mobile Radio Communication (PIMRC 2012), , Sydney, Autralia, Set htt://dx.doi.org/ /pimrc [32] P. Kolio, V. Frideriko and K. Paadaki, Ultra low energy tore-carry and forward relaying within the cell, IEEE Vehicular Technology Conference,. 1-5, Set htt://dx.doi.org/ /vetecf Coyright Diclaimer Coyright reerved by the author(). Thi article i an oen-acce article ditributed under the term and condition of the Creative Common Attribution licene (htt://creativecommon.org/licene/by/3.0/). 49