Priority-based Spectrum Allocation for Cognitive

Priority-based Spectrum Allocation for Cognitive Radio Networks Employing NC-OFDM Transmission Chittabrata Ghosh1 , Si Chen2 , Dharma P. Agrawal1 , and Alexander M. Wyglinski2 , 1

Department of Computer Science, University of Cincinnati, Cincinnati, OH, 45221 USA. 2 Department of Electrical and Computer Engineering, Worcester Polytechnic Institute, Worcester, MA, 01609-2280 USA. E-mail: {ghoshc, dpa}@cs.uc.edu, {chensi, alexw}@ece.wpi.edu.

Abstract—In this paper, we present three novel priority-based spectrum allocation techniques for enabling dynamic spectrum access (DSA) networking for non-contiguous orthogonal frequency division multiplexing (NC-OFDM) transmission. With each communication link in the network possessing a specified pair of bit error rate (BER) and throughput requirements for supporting a specific application, the proposed technique assigns one or more blocks of wireless spectrum to these applications in an attempt to simultaneously satisfy these requirements. Specifically, the proposed techniques assigns blocks of spectrum possessing aggregate bandwidth that is sufficient for supporting the intended wireless data service over the communication link. Moreover, since several portions of the wireless spectrum may be heavily attenuated due to frequency-selective fading resulting from multipath propagation, communication links requiring high error robustness are assigned frequency bands located further away from these attenuated regions of spectrum. Thus, the proposed spectrum allocation techniques aims at accommodating communication links supporting several wireless services possessing different performance requirements.

I. I With the recent surge in demand for broadband services on mobile devices, the problem of bandwidth scarcity became an emerging issue for the wireless industry. Cognitive radio [1] seems to be a panacea which facilitates sharing of licensed spectrum, such as television bands for example, among the unlicensed users. It is essential to sense the licensed spectrum over time for unused regions, also called white spaces, before being used by the unlicensed users, also referred to as secondary users (SUs). Since the utilization of the spectrum by the licensed users, also called primary users (PUs), is indeterministic, sensing of white spaces become challenging. Additionally, the PUs may even transmit only over a section of its licensed band which may lead to smaller unused blocks over the entire spectrum. The SUs need to employ special techniques to coexist with the PUs in these bands. This coexistence is also known as underlay spectrum sharing [2]. Given that This work has been submitted in parts to the Elsevier Computer Communications - Special Issue on Cognitive Radio and Dynamic Spectrum Sharing Systems [10].

channel conditions vary over frequency, and that performance demands of the wireless nodes within the network might be different, it is important to intelligently assign spectrum bands to various transmissions based on their requirements, especially for multiuser orthogonal frequency division multiplexing (OFDM). This is called spectrum allocation [3]–[6], [8]. In our paper, we have exploited the information available from the spectrum occupancy data [10] to gauge the spectrum utilization of the spectrum at any time instant. Additionally, we have exploited the idea of non-contiguous OFDM (NCOFDM) for higher spectral efficiency and spectrum sharing among prioritized SUs. NC-OFDM, a concept illustrated in details in [11], allows coexistence of SUs and PUs in the same operating sub-bands. Finally, we propose and compare three novel priority-based spectrum allocation schemes, namely: First Available First Allocate (FAFA), Best Available Selective Allocate (BASA) and Best Available Multiple Allocate (BAMA) based on the throughput and bandwidth utilization metrics. The major contributions of our paper are: •

•

•

•

Define priorities among SUs based on the BER, throughput, and sustainable delay constraints; Computation of active blocks in sub-bands of an operating spectrum from our spectrum occupancy model validated on real-time measurements from paging (928-948 MHz) bands; Estimate BER supported on thess active blocks in subbands based on the fading channel coefficients; and Implement FAFA, BASA and BAMA schemes for priority-based spectrum allocation among prioritized SUs.

The rest of the paper is organized as follows. Section II discusses the related work. Section III briefs the spectrum occupancy model which facilitates our priority-based spectrum allocation algorithm. Section IV illustrates our proposed algorithm using NC-OFDM and channel state information. Section V deals with the implementation of our proposed algorithms and the performance evaluation of our proposed scheme. Finally, Section VI draws the conclusion.

II. S M Cognitive radios are capable of spectrum sensing, which is essential in order to avoid interference with the PUs. Present spectrum occupancy information, together with spectrum occupancy statistics, becomes an important part of spectrum allocation. Specifically, spectrum block with longer mean idle time are assigned to requests for more durable transmissions. Moreover, attenuation profile can be utilized to assign frequency bands with less attenuation to communication links requiring high error robustness. In this work, we consider a network with one base station and numerous cognitive radio users. The users generate M applications represented as iapp , i = 1, 2, · · · , M as shown in Figure 1 and requests for sub-bands in the operating spectrum. The users are within one-hop transmission range of the base station. Each request is received at the base station, which assigns a priority value based on the ith app user BER, delay, and throughput requirements. The base station also detects the active blocks (information extracted from the above explained spectrum occupancy model) in the N sub-bands of the operating spectrum. Each sub-band is represented by jsb , j = 1, 2, · · · , N. Based on the size of the detected block in a sub-band, the base station computes the number of active sub-carriers (P − m) in the jth sb sub-band, where m represents the number of inactive sub-carriers in an NC-OFDM symbol of total P sub-carriers. This process is iterated for all the sub-bands 1, 2, · · · , N in the operating spectrum. The base station also computes the BER supported by the jth sb sub-band while taking fading channel coefficients into consideration. Thus, as shown in Figure 1, the base station maintain a database of BER and active sub-carrier values for all subbands. This database needs to be updated with time, using the PU occupancy statistics obtained from the spectrum occupancy model. Now, the base station maps a suitable sub-band jsb to a request iapp , satisfying its priority derived from the BER, and delay requirements. The channel allocation table maintains the recent status of priority-based spectrum allocation performed by the base station to satisfy requests from M SUs. Note that an application is queued if no sub-band can be scheduled to satisfy its requirements. The details of our proposed schemes are illustrated in the next section. III. P P- S A

 NC-OFDM  C I The cognitive radio user requests are queued at the base station for probable service. The user requests can be prioritized based on throughput, BER, and sustainable delay requirements. For example, a real-time application can sustain delay less than or equal to 40 ms with BER requirement of 10−7 , while a typical high quality audio application can sustain a maximum of 200 ms delay over a channel with BER of 10−5 [12]. A. NC-OFDM Sub-carriers for Priority NC-OFDM allows usage of idle non-contiguous sections of the sub-band by de-activating those sub-carriers of OFDM

Figure 1. Schematic diagram of the system model used for proposed prioritybased spectrum allocation techniques among SUs

present in the sub-bands being used by the PUs. Let the bandwidth assigned for an NC-OFDM symbol with P subcarriers is denoted by B Hz. Hence the bandwidth for a single sub-carrier of NC-OFDM is (B/P) Hz. For illustration purposes, let us consider the TV bands, each of which occupies 6 MHz bandwidth. If m sub-carriers are made inactive for PU occupancy, then the loss of bandwidth can be expressed as (B/P) × m. Hence the effective bandwidth EBw for data transmission in an NCOFDM system can be written as: EBw = B − (B/P) × m = B[1 − m/P].

(1)

Therefore, the effective capacity EC of an NC-OFDM symbol can be written as: EC = EBw log(1 + S NR) = (1 − m/P) × C,

(2)

where C is the Shannon capacity of the NC-OFDM system and SNR is the signal-to-noise ratio of the NC-OFDM symbol. Observe that 0 ≤ m/P ≤ 1 for m = 0, · · · , P. Eq. (2) implies that the fraction m/P plays a decisive role in defining the effective capacity. In other words for a fixed value of P, smaller value of m, i.e., few inactive sub-carriers in a block of subbands, yields a higher EC when compared to higher values of m. This idea leads to the core concept of our unique spectrum allocation strategy using NC-OFDM. This is explained in the following sub-section. 1) Channel Allocation using NC-OFDM: Each NC-OFDM symbol consists of P narrow-band sub-carriers divided over a fixed bandwidth B. This bandwidth can span across a definite number of sub-bands in the spectrum considered. Hence it is essential to have the knowledge of the following: •

•

PU occupancy in the sub-bands considered for NCOFDM; bandwidth occupied by each PU in these sub-bands.

Once we have this information available, the cognitive base station can decide on the number of sub-carriers to be made inactive. The following sub-section illustrates on obtaining this value m.

2) Number of inactive sub-carriers in NC-OFDM: Let us assume that the ith PU occupies a bandwidth BWi in the ith sub-band, where i = 0, 1, · · · , N. The number of sub-carriers i of NC-OFDM in the ith sub-band is BW B/P . Hence, the number of inactive sub-carriers m can be expressed as: m=

N X i=1

BWi

P B

= (NP)

!

BWequal , B

(3)

Application 2

Secondary users could have different capability of transmission power based on their battery energy left, while frequency selective channel fading causes different transmission power requirement in different bands. Such consideration of transmission power increases the complexity of band allocation. Before associate a secondary user application to a band, first we have to check whether the secondary user can provide enough power to satisfy the power requirement of that particular band. C. Channel State Information The channel state information is critical for better service quality among the SUs. The channel state information is utilized in our priority-based spectrum allocation scheme in the form of BER. Therefore, a channel experiencing flat fading or no fading can support higher BER applications while a channel with deep or fast fading can support low BER services. In our system model, in addition to the additive white gaussian noise (AWGN), we have introduced fading channel coefficients in each channel. The fading coefficients dictates the channel quality at any instant of time. IV. S A S In this paper, we present three different spectrum allocation strategies based on the following considerations: • NC-OFDM sub-carriers and BER parameters to define preferred sub-bands; and • Priorities among applications generated from the secondary users based on throughput, arrival time, BER, and delay requirements. A. First Available First Allocate (FAFA) In this proposed scheme, the cognitive base station constructs a priority queue based on the arrival time of each application into the queue. This strategy follows the spectrum allocation technique illustrated in [7]. FAFA selects the first application from the queue and allocates the first available sub-band that satisfies the BER requirement of the application. The scheme follows the same strategy for the subsequent subbands. This type of spectrum allocation does not compare

Cognitive Base Station

Constructs priority queue based on BER and delay constraints

Compute BERsb (j) and msb (j) for j-th sub-band, (j = 1, 2, . . ., N)

Application M For i = 1:M N

Y

Is BERsb(j) == BERi || BERsb(j) == BERi / 10 ?

For j = 1:N Is alloc_chnj == 0 && alloc_appi == 0?

Next j N Next j

Y Next i

where we assume that all the PUs occupy the same bandwidth BWequal in their respective sub-bands. Hence, the cognitive node can compute the effective bandwidth EC by substituting value of m in Eq. (2) as: ! BWequal C. (4) EC = 1 − N B B. Variable Transmission Power Requirements

Receives BERi and ni for i-th application, (i = 1, 2, . . ., M)

Application 1

Store j-th sub-band with BERsb(j) == BERi || BERsb(j) == BERi / 10

Select a sub-band out of stored sub-bands with min msb (j)

Figure 2.

Allocate sub-band j to application i, alloc_appi = 1, alloc_chnj = 1

Flow diagram of the BASA priority-based spectrum allocation

BER requirements of other applications in the queue before allocating a sub-band to its requesting application. B. Best Available Selective Allocate (BASA) In this scheme, the cognitive base station constructs a priority queue based on the joint requirement of BER and delay requirements. It is critical to comment at this point that all applications share a common relation between BER and delay requirements. All real time applications (implying better BER performance) such as streaming media, video conference, online video games, etc. are delay sensitive. Hence, they can sustain only a minimum transmission delay (< 40 ms). Similarly, all non-real time applications, such as data applications, are robust against fading channel and hence can be served on sub-bands with lower BER performance. Additionally, these non-real time applications can sustain delays in the order of 200 to 400 ms. This priority-based spectrum allocation scheme does not consider throughput in assigning priorities along with BER and delay constraints. The reason for this consideration is that there are many typical applications. For example, applications from public safety require less bandwidth and better BER performance. Similarly, there are applications, such as example high quality audio requiring 940 kb/s throughput and BER of 10−5 , that demand higher bandwidth while simultaneously robust against fading channels. Therefore, to maintain fairness among the requesting applications, priorities are defined based on only delay and BER constraints. In BASA, the cognitive base station allocates the best available sub-band that satisfies the selective requirement of a SU. This selective allocation is based on the prioritized BER and delay requirements and as well as the throughput, i.e., bandwidth demand. The flow diagram of BASA is depicted in Figure 2. On receiving M requests from the SUs, the base station prioritizes an ith request, i = 1, 2, · · · , M, into a queue on the basis of its BER and sustainable delay requirements. It also computes the BER support BERsb ( j) and the number of inactive subcarriers of NC-OFDM symbol msb ( j) for each jth sub-band

C. Best Available Multiple Allocate (BAMA) This multiple access technique has an initial operation similar to that of BASA described above. BAMA performs its spectrum allocation in two stages as follows: •

•

Allocation of prioritized requests to selective sub-bands similar to BASA; and Detects un-allocated user requests and schedules them to previously allocated sub-bands, which satisfies the BER and throughput requirements.

Hence, the flow diagram in Figure 2 also applies for the execution of the first stage in BAMA. The potential of BAMA over BASA is reflected with the following step of multiple access using NC-OFDMA. For each ith SU request in the priority queue, the cognitive base station checks the ith index of the vector alloc appi for an un-allocated request. For each unallocated request in the priority queue, it searches and stores sub-bands with equal or one order lower in magnitude of BER support (i.e., BERsb ( j) == BERi or BERsb ( j) == BERi × 10). The reasons for including a checking condition for a sub-band with lower order BER are as follows: •

•

Possibility of absence of a sub-band, even during the first stage of operation, with equal order of BER support BERsb ( j) as that of BERi ; or Possibility of consuming all the sub-bands of equal order of BER support as that of BERi during the first stage of execution of BAMA, to requests higher in priority than this un-allocated user request.

Thus, instead of queuing the packet for the next time instant, the base station makes an attempt to allocate the waiting request to a sub-band, only if, its BER support is of equal or with one order lower in magnitude as that of the BER requirement of the requesting user. In the final stage, the base station allocates the previously allocated sub-band with maximum number of remaining un-allocated active subcarriers (i.e. max rem act subcar j ). This supports the name of our spectrum allocation scheme since the base station allocates the best (i.e., the sub-band with maximum number of un-allocated active sub-carriers) to an un-allocated user request.

Proportion of active sub-carriers

using Eq. (3), j = 1, 2, · · · , N. The two vectors, alloc chn j and alloc appi are as defined for FAFA, respectively. Similar decision, as described in FAFA, is performed in the first decision block. The improvement in BASA over FAFA evolves from the following decision blocks. Those sub-bands with BER support BERsb ( j) equal to or one order better than the requested BER BERi of the ith SU are stored. Out of these stored sub-bands, a sub-band with the highest number of active sub-carriers (i.e., minimum msb ( j)) is allocated as illustrated in Figure 2. Finally, the status of the jth sub-band and ith SU request are set appropriately to prevent duplicate allocation of sub-bands and user requests (i.e., alloc chn j == 1 and alloc appi == 1). This process is iterated over all the M requests.

Sub-band numbers Figure 3. Proportion of active sub-carriers for NC-OFDM for all sub-bands for ten time instants

The unique characteristic of BAMA is that it allows flexibility to the user requests with variations in throughput requirements. The priorities are defined based on BER and delay requirements of the user requests. Now, for user with lower throughput but higher BER constraint will be treated with same priority as that of an user request with higher throughput and similar BER requirements. This allows fairness among the users from the perspective of throughput constraints. V. S R In this section, we present the performance evaluation of each of the three proposed schemes namely, FAFA, BASA, and BAMA. The schemes are evaluated in terms of the bandwidth utilization and throughput achieved for equal number of SUs requests arriving at the base station. The allocation efficiency is evaluated in terms of the number of un-allocated requests per allocation time slot. The simulation is run for one complete duration of 1000ms and each allocation slot is allocated 100ms. Hence, within the complete duration, there are 10 allocation slots. Within a allocation period, we assume that the PU occupancy status remains unchanged once they have arrived in their respective sub-bands. This assumption is valid since that the history of spectrum occupancy dictates how we allocate our transmission frequencies, which means spectrum that has been occupied before has a higher chance of being occupied again compared to spectrum that has not been occupied so far. Thus the parameters (i.e., number of inactive sub-carriers m as well as the active sub-carriers (P−m)) utilized in our allocation policy will remain steady compared to the rate of secondary user requests. For our simulation, the term “active sub-carriers” gives a total of all contiguous and noncontiguous sub-carriers in a sub-band. The simulation was conducted on a spectrum bandwidth of 102 MHz with sub-bands of 6 MHz bandwidth. Hence, we have 17 sub-bands in our simulated spectrum. Each NCOFDM symbol is implemented with 256 sub-carriers. Finally, we have considered that each sub-band is licensed to only 1 PU. Hence, we have 17 PUs occupying each of the 17 subbands at some particular time in the operating spectrum. One of the critical parameters for our spectrum allocation

while BAMA still has the same performance. In the result, the performance of FAFA and BASA are almost the same and their performance curves overlap in Fig. 4, which means the first available band is almost always the best available band. This is because in TV broadcasting channels, primary users either occupy the channel almost entirely or not at all. Thus, first available allocation scheme is the same as best available allocation scheme in single allocation cases.

450 400 350 FAFA without power constraint 300

BASA without power constraint

250

BAMA without power constraint

200

BASA with power constraint BAMA with power constraint

150

FAFA with power constraint

100 50 0

1

Figure 4.

2

3

4

5

6

7

8

9

10

Average throughput comparison of FAFA, BASA, and BAMA

is the number of active sub-carriers (P − m) in each sub-band of the simulated spectrum. Fig. 3 refers to (P − m) for each of the 17 sub-bands during a simulation period of 1000ms, consisting of 10 instants each of 100ms allocation duration. Therefore, as shown in Fig. 3, during the first time instant subbands 1, 3, 7 - 9, and 16 have idle bandwidth equivalent to 256 active sub-carriers, indicating that these sub-bands are idle for that time duration. Sub-band 13 is least preferred during the first time instant since it has 34 active sub-carriers. Over the entire simulation duration, it is noted that sub-bands 1, 3, 7 9, and 16 have maximum number of active sub-carriers, i.e. 256 sub-carriers over all 10 time instants. Sub-band 5 has the least number of active sub-carriers, i.e., a total of 912 over all the 10 instants. This suggests that sub-band 5 is heavily used by a PU during the entire simulation duration. Hence, combining information retrieved from Fig. 3 and the BER result from all the subbands, it is concluded that higher priority application requests with BER of the order of 0.001 are queued during the first time instant because of inavailability of suitable sub-bands. On the contrary, during the second time instant, sub-band 9 is best preferred with 256 active sub-carriers and BER support of 0.00028. Since FAFA and BASA allocation schemes allocate one SU per subband, only one higher priority request will be served on subband 9 while other higher priority requests have to be queued. The BAMA scheme, being a multiple access scheme, will allocate sub-band 9 to its capacity of 256 active sub-carriers to a number of high priority SU requests. This explicitly explains higher bandwidth and better throughput performance of BAMA over FAFA and BASA. We then added power constraint to the senario to test the three schemes. The power constraint is set such that in those bands occupied by primary users, allowed transmission power for secondary user is only -60dB. And secondary users are assigned transmission power level randomly such that there is a chance of 20% that the secondary user does not have enough transmission power to establish successful connection. This setting is to simulate the case with selective channel fading. Fig. 4 shows the average performance of the three schemes over 50 simulations. With transmission power constraint, FAFA and BASA suffer great performance degradation,

VI. C In this paper, we have developed three unique spectrum allocation techniques to satisfy requests generated from secondary users (SUs). Our simulation results prove the effectiveness of using NC-OFDM along with BER support information before sub-band allocation to SUs. With additional information of active sub-carriers in NC-OFDM and non-contiguous Orthogonal Frequency Division Multiple Access (NC-OFDMA) facilitates in achieving higher throughput and better bandwidth utilization. We have also proved that our proposed schemes performs much better than traditional OFDM approach with respect to the rate of spectrum allocation to spontaneous SU requests. A The authors would like to thank the University Research Council (URC) Graduate Student Research Fellowship Program at the University of Cincinnati for generously supporting this research project. R [1] J. Mitola III, Cognitive radio: An Integrated Agent Architecture for Software Defined Radio. Ph.D. Thesis, KTH, Stockholm, Sweden, 2000. [2] Q. Zhao and A. Swami, “A decision-theoretic framework for opportunistic spectrum access,” IEEE Wireless Comm. Mag., vol. 14, no. 4, pp. 14-20, August 2007. [3] C. Y. Wong, R. S. Cheng, K. B. Letaief, and R. D. Murch, “Multiuser OFDM with adaptive subcarrier, bit, and power allocation,” IEEE Jour. Selected Areas in Commun., vol. 17, no. 10, Oct. 1999, pp. 1747-1758. [4] F-S. Chu and K-C. Chen, “Fair adaptive radio resource allocation of mobile OFDMA,” IEEE Intl. Symp. on Personal, Indoor and Mobile Radio Commun., PIMRC, 2006, pp. 1-5. [5] P. Zhu, J. Li, and X. Wang, “Scheduling model for cognitive radio,” Proc. Int. Conf. Cognitive Radio Oriented Networks and Commun.,2008 pp. 1-6 [6] R. Urgaonkar and J. Neely, “Opportunistic scheduling with reliability guarantees in cognitive radio networks,” IEEE INFOCOM, 2008, pp. 1975-1983. [7] T. A. Weiss and F. K. Jondral, “Spectrum pooling: an innovative strategy for the enhancement of spectrum efficiency,” IEEE Commun. Mag., Vol. 42, Issue 3, March 2004, pp. S8 - 14. [8] A. Attar, O. Holland, M. R. Nakhai, and A. H. Aghvami, “Interferencelimited resource allocation for cognitive radio in orthogonal frequency division multiplexing networks,” IET Jour. Institute of Engineering and Technol., vol. 2, no. 6, 2008, pp. 806-814. [9] F-S. Chu and K-C. Chen, “Radio resource allocation in OFDMA cognitive radio systems,” Proc. 18th Annual IEEE Int. Symp. Personal, Indoor and Mobile Radio Commun., 2007, pp. 1-5. [10] C. Ghosh, D. P. Agrawal, S. Pagadarai, and A. M. Wyglinski, “Statistical spectrum occupancy modeling employing radio frequency measurements,” submitted to Elsevier Computer Commun., 2008. [11] R. Rajbanshi, A. M. Wyglinski, and G. J. Minden, “An efficient implementation of NC-OFDM transceivers for cognitive radios,” Proc. 1st Int. Conf. Cognitive Radio Oriented Wireless Networks and Commun., Mykonos Island, Greece, June 2006. [12] G. Aggelou, Mobile Ad Hoc Networks. McGraw-Hill Pub., 2005.