Approach* unused. describes our scheme, Sensor based Dynamic Channel

Transcription

1 Wireless Sensor based Dynamic Channel Selection in Cellular Communication by Cognitive Radio Approach* orsha Banerjee, Student Member, IEEE, Chittabrata Ghosh, and Dharma P. Agrawal, Fellow, IEEE University of Cincinnati, Cincinnati, OH 41- OBR Center for Dtributed and Mobile Computing {banerjta, ghoshc, Abstract. In a cellular communication scenario, wireless sensors can be deployed to sense the interference power of the frequency band. In an ideal channel, interference temperature (I) which directly proportional to the interference power can be assumed to vary spatially with the frequency of the sub channel. We propose a scheme for approximating Is over an extended C-band (licensed and unused televion band) by fitting sub channel frequencies and corresponding Is to a regression model. Using th model, I of a random sub channel can be calculated by the base station (BS) for further analys of the channel interference. Our proposed model based on readings reported by Sensors helps in Dynamic Channel Selection (S-DCS) in extended C-bandfor assignment to unlicensed secondary users. S-DCS maximizes channel utilization and proves to be economic from energy consumption point of view. It also exhibits substantial amount of accuracy with error bound within.oo. Again, users are assigned empty sub channels without actually probing them, incurring minimum delay in the process. Overall channel allocation efficiency also maximized along with fairness to individual users. Keywords: channel allocation, cognitive radio, interference temperature, regression, WFQ I. INRODUCION Static channel allocation of licensed bands to the primary users results in wastage of channel bandwidth and in turn, lowers spectrum utilization. h problem can be taken care of channels. Co-extence of primary users (PUs) and SUs limits the usage of only one data channel by a SU at any time in spite of the availability of several adjacent channels. For efficient spectrum utilization, the cognitive radio should be able to sense the idle channels and allocate them dynamically to SUs on-demand. herefore, the SU (or the cellular users) should be able to use multiple channels simultaneously or else could combine several parts of the bandwidth into a single wider channel. Maximizing the utilization of unused spectrum has been a key sue and various ideas have been proposed in literature ([1], []) for dynamic channel management. For improved spectrum utilization, a set of sub-channels in an extended C-band can be assigned to SUs when the subchannels are not utilized by the PUs. A cellular BS dee. Atahes Again, p od regres mode,can imleentingeou unused sub-channels and assign them to incoming SUs. SUsatbhein ulcneusroftseubchannels,asatvthey.sncee tos P have higher priority over the secondary users, the former can use the entire C-band whenever necessary. It the job of the SU to detect the arrival of the PU (also called spectrum sensing) and leave the sub-channel corresponding to that Cband of the recent participating PU. Otherwe, the power level of the PU's signal being much higher than that of the SU, the former will entirely corrupt the signal of the latter resulting in interference, called the Primary User Interference (PUI). Again, when an SU uses a particular sub-channel, signals from the adjacent SUs can interfere with th desired user. h interference has been termed as adjacent channel interference, ACI. Since, ACI can result in substantial packet loss, it necessary to obtain a measure of th interference. Our scheme strives to minime th ACI with farthest channel assignment. he degree of noe level introduced because of ACI measured in terms of interference temperature (I). For an ideal sub-channel, I assumed to vary spatially with the carrier frequency. herefore, ACI can be modeled as a function of the sub-channel. he rest of th paper organized as follows: Section II lts the preliminaries and dcusses exting work. Section III describes our scheme, Sensor based Dynamic Channel Selection (S-DCS) in details. Simulation results are presented in Section IV. Finally Section Vconcludes the paper. by channel sensing and dynamic allocation of the licensed bands, performed by specially designed radio called cognitive radio. Cognitive radio a revolutionary technology that primarily aims for substantial spectrum efficiency with the aid of advanced spectrum sensing and dynamic channel assignment in licensed bands without actually obtaining a license [1]. It built on a software defined radio [1] and capable enough to sense its surrounding environment and take decions based on that. Opportuntic spectrum sharing of the licensed spectrum [4] by secondary users (SUs) one of the key conlcepts being explored inl recenlt research scenario. he colntrol messages are sent over a common control channel whie data transfer takesplace overa numberof data * h work has heeln supported hy the Ohio Board of Reagenats' Doetoral Enahalneemelnt Funlds //$ IEEE ona

2 IL. RELAED WORK hopping from one frequency to another with time. he idea of how I can provide a measurement of spectrum holes i.e. idle sub-channels in a frequency band proposed in [1]. In S-DCS, I of a sub-channel compared with a prespecified threshold, h (which adaptive with the degree of activity in the band) and if it <th, it implies the sub channel empty and can be assigned to an incoming SU. Sai Shankar et al. [] proposes a collocated architecture for cognitive radio for efficient spectrum sensing. It defines two networks, one exclusively for spectrum sensing and the other an operational network. S-DCS can be looked upon as providing a solution for spectrum sensing and operational network where sensors do spectrum sensing with the operational function implemented at the BS, thus providing the functioning of a cognitive radio. In [] architectures for spectrum sensing with wireless sensor network are proposed which might or might not involve an opportunity identifier (to determine whether a particular band can be used or not) or an opportunity manager (controller which controls all the design specifications) depending on the type of application. Zheng et al. [] proposed a scheme for spectrum usage among a set of PUs and SUs. S-DCS focuses on dynamic spectrum allocation among only SUs. A key sue has been brought UP in [] that direct communication of users with low energy devices, like sensors, may lower the sensor's energy even further. hus, a centralized authority needed for communication. In S-DCS, a BS (having much more energy than sensors with low battery power) the interface of communication between the users and the sensors thus saving a lot of overhead of managing the sensors. a. Proposed Regression modelfor channel sensing During the time frame, BS selects a random set of sensors (SS or Selected Sensors) to report the corresponding I s to it. Each sensor selected by the BS for reporting reports the frequency its currently tuned to and the corresponding I. Our channel conditions being assumed ideal, Is of adjacent channels are assumed to be highly correlated and are thus III. OUR PROPOSED S-DCS SCHEME We consider dynamic channel allocation in cellular communication with a frequency band (extended C band) of. GHz-.7GHz []. he band assumed to be already in use by few SUs hence making the power levels very high. As illustrated in Fig. 1, BS of a cell chooses some sensors randomly for reporting their sensed I values during a time frame,. During th time, some users (three of them in Fig. 1) can request channels from the BS. Again, I directly proportional to the power level (P) and given by the following expression: (1) I=P+ (-44.7)/K.B Where K=Boltzmann constant=1. X 1-, B=bandwidth of the sub-channel, PAInterference power ( dbw/hz as defined in []), dbw/hz the allowable interference at ea t surface []. Hence, with higher P, I of the corresponding sub-channel also much higher than the h, giving a measure of how busy the channel. Wireless sensors are deployed within 1-hop of BS tuned to frequencies of the C-band. During a particular timne framne, a single sensor tuned to particular frequency and senses the I of that particular sub channel and keeps D-band I<< El II>< e es Sensortuned to a particular frequency of C-band User requesting channel frm B S Fig. 1. Cellular Network architecture for S-DCS modeled as function of frequency. he BS made cognitive by implementing a regression model for calculating the Is of all the sub-channels of the C-band based on the repored values thess. herefore, profrth SSlynmherefgr,ebsedionsmlrthe ore prport edvales rb ba of n the reportedvalues to ourbroosdior pefrmnoynma[rgesinsiia in [] to obtain f (x) = A + fi.x + fx() where x the frequency of a sub channel, f(x) the corresponding I,, 's are the model coefficients []. Eq. (1) gives a polynomial of degree, i.e., quadratic polynomial. As the degree of f(x) increases, the number of coefficients, i.e., keeps increasing []. With increase of degree, the accuracy of approximation keeps increasing and for our S-DCS scheme, we choose f(x) to be cubic as increase in number of coefficients increases computational complexity. A cubic polynomial provides a trade off between the accuracy and the computational delay. Again, only few sensors are needed to report their sensed data to the BS and their roles can be exchanged at different time frames thus providing increased energy savings of sensors. If an incoming user requests a subchannel from the BS, the BS first scans through the Is of each of the SS (by substituting the sub- NE D at one end l not at one end Fig.. Different scenarios for channel allocation

3 () 1 1 Fig.. Free Channel assignment to SUs with ±1 consideration channel frequency in the polynomial and getting the correspodn I tose if th I of a su-chne on < h. h done to determine if any of the sub-channels can be allocated. If not, then the BS finds a new sub channel for th user to avoid ACI. If none of SS can be allocated, BS needs to select some sub-channel whose data not directly reported by the corresponding sensor to the BS. h information can obtained by the BS by substituting the by... a sub.-chnnl. mesue in: terms o:fi Sinc, impac of I of neighboring sub-channels on i decreases proportionately with dtance, we need to consider a substantial number of neighboring sub-channels of i, and calculate the I of i based on the Is of these sub-channels. From simulation we find that two sub-channels adjacent to i (th scheme being called ± Iconsideration as sub-channels considered are i-i and i+1) i are sufficient to consider the final value of I of i and the total number allocated with ± 1 consideration given in Fig.. We also observe that channel allocation efficiency (total Algorithm: CHANNEL ALLOC (i) number of free channels assigned to incoming users among the total number of free channels in the band) with begin ±1L consideration higher than with ± consideration Case Fig (a). if Ii 1 (or Ii-I when i the last sub-channel) << (checking the ACIs of (i-i)th, (i± 1), (i-), (i+)h subchannels before allocating the ith sub-channel). Of all the free channels shown in Fig., channels are assigned with th... ± 1consideration unlike ± consideration which would give endif~ (r weni sth ls sb-hanl) 19 channels. channels also happens to be the total I,only number allocated by S-DCS to all the users bh arriving in the interval, as found from simulation results in set to>> Sec. IV. ±1consideration again has two cases to consider endif on the location of the sub channel, i in the C-band as based Case Fig (). if (li << i allocated to user and Iiti th set > illustrated in Fig. and the cases are handled in our proposed i. alcieated so user a -i Ii else I.. endifi s elseif (Ii+I << se set > to,, < t < ) i allocated to user and Ii set > algorithm, CHANNEL_ALLOC. b (Ii-I and, elseif (I> and I> 1S <IX ) I allocated to user and IIi-I tb,elseif 1 set > th < else Ii set to < endif end endif endif channel IDs (considered to be unique) of each sub-channel sequentially in f(x) to get the corresponding I and compare it with h. Whether or not to allocate a sub-channel i (whose I <th) to an incoming user or not also based on the ACI offered by sub-channels adjacent to i and therefore th decion making procedure described in the next subsection. b. Chalnnel interference alllocaltion minimizing aldjacent chalnnel If the I of a sub channel i found to be less than,rti it not allocated immediately to the requesting user as there an sueof miniming ACI that needs to be consied here Whetheriuitabletobe allocatedtoasubasedonthe Is or power lev o the adjacen cannels oo ecauseo e effect of ACI offered by these neighbors of i to i. ACI offered CHANNEL ALLOC illustrates how to allocate i to an incoming user ensuring that it idle and its assignment to an incoming user not going to affect the ACI of the adjacent channels badly. If most the adjacent sub channels are not free, it implies that users are already there in those sub channels therefore Ii set to > b so that no user given th channel as th user (if it allocated i) going to experience high ACI from adjacent channels because of exting users in the adjacent sub-channels. Due to the same reason, right after sub channel 7 assigned to an SU in Fig., though 77 the next free sub-channel in sequence that could be assigned, it not done as it would increase the ACI experienced by the SU in 7th sub-channel even further. If majority of the channels adjacent to i are free (given by their Is < ), then i allocated to an incoming user and its current Ii set to > b. thhese new readings of I are sent by the reporting sensors in the next period update. henew coefficients Eq. 1 when are updated changes. ofhus a new to of reflect these accordingly ' user requests a sub-channel, if the BS substitutes xi in f(x), the corresponding Ii found to be aloca ed to th asnewhuser. >,. and therefore not c. Fatirness to users usin?g a hyb1rid queuin?g shtrategy

4 In S-DCS, incoming users arrive randomly and are stored in a finite queue of size and we assume that the user can request channels more than once from the BS. Again, one user's request of sub channels can vary a lot from some other user's. If a strategy used throughout (termed pure ) for allocating channels to these users, then only very few users will have all their channel requests fully satfied and the rest will have zero channels assigned to them as all free subchannels would have been used up in meeting the requests of the first few users (no matter how many they request). herefore, we introduce a hybrid strategy [termed as Hybrid (i.e. +WFQ)], where a allocation strategy maintained until the total number of channel requests (by users in the queue) equals numberof free channels- offse. he offset a predefined system parameter needed to meet the condition. Once th strategy happens, weighted fair queuing (WFQ) followed. he more the number of sub-channels requested by a user and the more the frequency of arrival of the same user, its weight will be reduced. Based on th model, the number of sub-channels allocated given by: () r xf n c x i Wtot x -1 where ci= number that are allocated to user i by.. the BS. Wtt>= numberof freechannels- offse ni-total number of sub-channels in the C-band that are currently free ri -total number sub-channels requested by user i Wtot- itotal number of sub-channels requested by all the usersl who arrived before i and including i i.e.w = 1 i X. 4 rhj=l generated 1 1 Hurriber of serws reporfing to ink fi -frequency of arrival of user i within the time frame (users are the trade-off between communication energy and percentage error. () Variation in I (subsequently ACJ of offered by subchannels) with different types of channel allocation strategies. Fig. shows the condition of our extended C-band w.r.t. I when free channels are allocated using different types of channel allocation strategies. When adjacent free channels are allocated without any consideration to the impact of their ACIs on each other (called consideration), most of the channels exhibit very high I (Is of most of the sub-channels cross the flat line with th=.). With ±1consideration, users are allocated to those sub-channels whose immediate neighbors do not all have Is>th. hus, users experience less ACI as not all adjacent will have Is>th. We observe, similar results (even lesser ACI) with channel allocation with ± consideration following the same reasoning. However, for our model, we follow ±1L consideration as it achieves higher channel efficiency than with ± consideration as evident from Fig.. () Fairness to users with Pure and Hybrid F/FO. Fig. shows the variation in number channels allocated with our randomly in th time frame) Fig.4. Percentage in approximation varying the number of SS IV. SIMULAION RESULS We have simulated S-DCS scheme using C-. A MHz ~rnodi band (. GHz-.7 MHz) used with 1 sub- channels each.7 of bandwidth MHz. Once, BS has the polynomial approximating the Is of the band, users are generated. randomly every ms in the interval = secs. Users are then t.4 _. allocated free channels with ± 1 consideration. While. evaluating the results, we consider the following metrics: (1) Percentage error in approximating I using our proposed regression model by varying the number of SS: From Fig. 4...~~~~~~~~~~~~~~~~~~~~~ ; we observe that as number of SS to BS increases we observe that the percentage error E.K,,,q, Ipp' I1 x I I.,11 reporting in approximation of I using our model proposed regression dlecreases slighltly. Again, increasing numbher of SS reporting BS involves increases transmsion energy of sensors. herefore, S-DCS choose an optimal numbejr of 1 SS to report to BS following Channel - Ii before new channel assignment - Iaterchannelannianment with consideration Iaherchannelassignment with +-1 consideration Q* iafterchannel assignment ids of free subchannels with +- consideration Fig.. Minimizing ACI with varying considerations in channel selection

5 u 1o z * otal number of channels requested bya user X~~~~~~~~~~~~~EN*Io. Nallocated ocanels using pr ur No. of channels E1 Z using hybrid ~~~~~~~~~~~~~allocated V. CONCLUSION AND FUURE WORK ODs of users served by BS in time S-DCS achieves a stable margin of accuracy in approximating the I of a sub channel. It also provides substantial degree of energy savings by making only few sensors reporting data to BS. Since BS does not need to measure the I of each sub Fig.. Channel allocation efficiency of Pure vs. Hybrid _ No. allocated to user1 using pure... 4 ~~~~~~~No). ~ ~using ~ ~ ~ al ochybrid ated to user 1 u zno. No...t... requested by 1 No. of times of arrival of user user pure NoSSSSSSSSSSS11 allocated to using hybrid approximation scheme for fading channels. How S-DCS performs in presence of PUs and over multiple frequency bands also needs to be analyzed further. Switching time of SUs from one band to another because of arrival of a PU will influence the overall performance to a certain degree and therefore needs maj or consideration. REFERENCES of channel allocated to using channel individually, idle sub channels are allocated to users delay. Users experience reduced ACI due to farthest channel assignment implementation. Fairness to users also achieved not comproming on their individual channel requirements. Further study needed to apply our with minimum 7N.requested by user 1 offset), thus saving channels for meeting future channel requests. herefore, during its third arrival in the queue, user 1 gets some channels from the saved set thus being able to run its current application. hus our protocol provides fairness to all users by decreasing the priority of users the more number they request and the more frequently they request channels. [1] S. Haykin, "Cognitive... communicationst' IEEE Fig.7. No. allocated to repeating users using of Pure vs. Hybrid simulation, when 1 arrives for the nd time). After th instant, using pure, all free channels are exhausted after assignment to user, thus the subsequent users cannot not be assigned any more channels. he bars shaded black shows the total number requested by each user in the interval. he gray bar shows the number allocated to users using Pure. It closely follows the number of channels requested until wtot >-ni-offset (1-) no more channels can be assigned. Hybrid does fair channel allocation, by not meeting 1 of channel request of any user after whose arrival wtot >-ni-offset (user 1 gets instead of sub-channels; user gets instead of 4 sub-channels). hus, the channels saved by not doing th, are shared among the subsequent users according to their weights following Eq.. (4) No. allocated to of repeating users with Pure and Hybrid. If the same user requests channel from the BS more than once, pure meets all its channel requests until ni gets exhausted. In our simulation, users and 1 arrive more than once. In Fig. 7, with pure, we observe that 1 gets a11 the channels requested for the first two times it arrives. After that, it needed or channls bu didnot get any as n? got exhausted. hus, its last application cold not be run. With hybrid, 1 does not get all the channels requested when it arrives for the second time (as w~ >=~n- radio: brain-empowered wireless Journal Selected Areas in Communications, vol., no., pp. 1 -, Feb.. [] S. Nandagopal, C. Cordeiro, and K. Challapali, "Spectrum Agile Radios: Utilization and Sensing Architectures," Proceedings of IEEE DySPAN, Bazltimore, USA, Nov o 1 [] H. Zheng, and L. Cao, "Device-centric Spectrum Management," Proceedings of IEEE DySPAN, Baltimore, USA, Nov -]],. [4] L. Ma, X. Han and C-C Shen, "Dynamic open spectrum sharing MAC protocol for wireless ad hoc networks," Proceedings of IEEE DySPAN, Baltimore, USA, Nov I,. [] Federal Communications Commsion, FCC -9. [] orsha BaneiJee, Kaushik Choudhury, and Dharma P. Agrawal, "ree Based Data Aggregation in Sensor Networks Using Polynomial Regression," Proceedings of the th Ann. Conf on Information Fusion, July - 9, Philadelphia..