This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings Cross-layered Design of Spectrum Sensing and MAC for Opportunistic Spectrum Access Shoukang Zheng∗ , Ying-Chang Liang∗ , Pooi Yuen Kam† , and Anh Tuan Hoang∗ ∗ Institute for Infocomm Research, A*STAR Email: {skzheng, ycliang, athoang}@i2r.a-star.edu.sg † Dept of Electrical and Computer Engineering, National University of Singapore Email: py.kam@nus.edu.sg Abstract—In cognitive radio networks, the secondary users (SUs) are allowed to use the spectrum originally allocated to primary users (PUs) as long as the PUs are not using it temporarily This operation is called opportunistic spectrum access (OSA), and it is assisted through spectrum sensing In distributed OSA, the SUs sense the channel independently; once the channel is available, they contend for channel access on a frame-by-frame basis In this paper, we study the random medium access control (MAC) in conjunction with the sensing protocol design In particular, we are interested in the design of frame duration, sensing time and MAC random access to maximize the secondary network throughput performance while protecting the PUs from the interference of secondary users’ operations We formulate the nonlinear constrained optimization problems for the described system model with cross-layered and layered approaches Simulations show that the cross-layered approach performs much better than layered approach especially when the frame duration is small I I NTRODUCTION Recent research efforts on cognitive radio (CR) have opened the door to more efficient spectrum utilization [1] [2] In CR networks, the secondary users (SUs) are allowed to use the spectrum originally allocated to primary users (PUs) as long as the PUs are not using it temporarily This operation is called opportunistic spectrum access (OSA) To suppress the interference to the PUs, the SUs have to perform spectrum sensing before their attempts to transmit over the spectrum Upon detecting that the PU is idle, SUs can make use of the the spectrum for transmission so that the overall utilization efficiency of the spectrum is enhanced In centralized OSA, the SUs perform spectrum sensing in a cooperative manner Once the channel is found to be available, the user access is controlled in a centralized fashion Since sensing is imperfect, a false alarm reduces the opportunity to reuse the channel, while a misdetection impairs the performance of the PUs due to the SUs’ interference In [4], for a given frame duration, the trade-off between spectrum sensing time and SU’s throughput has been studied, and it is found that there exists an optimal sensing time which maximizes the throughput of the SU under the constraint that the detection probability is not less than the required protection threshold In [5], for a given sensing duration, the authors further optimize the frame duration under the constraint that the collision rate with the PU when it becomes active in the rest of the frame, is smaller than a certain threshold In distributed OSA, the SUs perform spectrum sensing independently and, once the channel is found to be available, each user will access it in a distributed manner [6] considers the joint PHY-MAC design of distributed OSA in a slotted structure of PUs with imperfect sensing as a constrained POMDP (partially observable Markovian decision process), and determines the optimal action with the policy following a seperation principle, based on the observation history The joint design approach of spectrum sensing and access is widely considered in the literature, e.g adaptive scheduling on the spectrum sensing periods [7], which is based on knowledge of channel conditions and maximizes the spectrum efficiency of CR operations for both cases of greedy transmission and stochastic arrivals on the time-varying channels [13] also presents two spectrum access schemes using different mechanisms to achieve secondary performance under primary constraints of collision probability and overlapping time It gives the closed form analysis on SU performance through a capacity bound and reveals the impact of various design options, such as sensing, packet length distribution, back-off time etc We study an OSA case in which some SUs get access to a licensed channel with a distributed sensing Prior to the transmission, SUs, i.e., CR nodes carry out sensing over the target spectrum for every frame The SU can then attempt transmission independently under a random access mechanism, if primary network is idle The spectrum access is decided separately without taking into account other users’ sensing result and/or decisions Our work is related to [8] Assuming the MAC traffic model of PUs and the probability distribution of SUs to be known for the design, our work mainly considers how to choose the optimal values of frame duration, sensing time and p-persistent probability for the problem of throughput maximization subject to a certain constraint on the SUs’ interference to the PUs In the layered design, the frame duration and sensing time are optimized by tradeoff of sensing-throughput over the time and independently, MAC layer throughput is maximized by selecting the best probability for p-persistent random access Cross-layered approach instead determines the three parameters jointly Both approaches take into consideration SU transmission interference to the primary network on the MAC layer The rest of the paper is organized as follows In Section 978-1-4244-2948-6/09/$25.00 ©2009 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings Fig SU frame structure with PU idle and busy periods II, the system model is described along with the assumptions made and Section III gives the formulation of the constrained optimization problem with triple parameters: frame duration, sensing time and p-persistent probability Numerical results and some discussion are provided in Section IV Finally we conclude the paper in Section V II SYSTEM MODEL A System Model We consider a primary network with the model as in [12] This established prediction on the semi-Markov model is approximated by CMTC (Continuous-time Markov chain) with a good fitting exponential distributions as idle and transmission (busy) periods Therefore in our paper, we follow the same CTMC model with exponential parameters λ and μ for idle and busy state, whose probability density functions are fi (t) = λexp(−λt) and fb (t) = μexp(−μt) respectively The stationary distribution of the idle and busy period are μ λ and λ+μ respectively λ+μ The frame structure of SUs are delineated in Fig.1 with which n SUs contend on the channel to transmit data to a destination through opportunistic utilization of a licensed channel The operation of such n users includes independent spectrum sensing before transmission and distributed channel access randomly n SUs operate on the frames, with each frame having an equal duration of T Assume all SUs can synchronize frame boundaries ideally and have to detect the availability of the channel prior to every transmission If there are no PUs communicating with each other, each SU may start to transmit The sensing time is also known to the destination so that it can synchronize with the starting of SU transmission B Spectrum Sensing When a primary network is active, the higher the SNR γ of the PU’s signal at the receiver of a CR device, more accurate it is to detect Pd and Pf are two important parameters for spectrum sensing, where Pd is probability of detections and Pf is probability of false alarm Pd is the probability that SU must accurately detect the presence of an active primary network The higher the value of Pd , the better the protection for primary operation According to one example set up in [3], the required detection probability is above 90% for cognitive radio systems operating on VHF/UHF TV bands Pf is the probability that SU falsely detects the presence of PUs when in fact none of them is active at the sensing time From the viewpoint of SUs, the lower the value of Pf , the higher the spectrum utilization Let τ be the duration that each SU carries out spectrum sensing, when Pd and SNR γ are fixed, given a sensing algorithm (e.g., energy detection or cyclostationary detection), Pf is a non-increasing function of τ [9]-[10] [4] assumes that the primary signal is independent and identically distributed (i.i.d), complex PSK modulated, with zero mean; the noise at CR receivers is circular symmetric complex Gaussian with zero mean; and the primary signal and noise are independent Hence, based on energy detection algorithm, the probability of false alarm is calculated by: Pf (τ ) = Q( 2γ + 1Q−1 (Pd ) + τ fs γ), (1) where fs is the channel sampling rate and Q(.) is the complementary distribution function of a standard Gaussian variable, i.e., ∞ √ e−t /2 dt (2) Q(x) = 2π x C MAC Random Access In the classical collision channel model, for each wireless link, we model the collision channel with three possible channel outcomes: idle, success and collsion that occur respectively if none, one or more than one packets transmit and simultaneously reach the same destination receiver at the same period of time Upon detecting the idle channel, each SU will decide independently whether to transmit on the channel There are different approaches to avoid collisions for contention, for example, the schemes such as random backoff and p-persistent In our study, we only consider the case when SU, upon detecting the idle channel, will attempt to transmit data using p-persistent random access with probability p Given p, the probability that there will be one successful transmission for n SUs contending one channel, when the primary network is idle, is: Psucc = np(1 − p)n−1 (3) The probability of at least one SU transmission is Ptr = − (1 − p)n (4) The collision not only occurs between SU transmissions with a random access manner, but also generates between the transmission of PUs and SUs because of the imperfect spectrum sensing When PUs are busy, misdetection on the channel in SUs leads to possible collision Even if SUs can perform sensing without misdetection, PU may transmit at any time provided no sensing can be done simultaneously with its transmission and SUs are not able to synchronize with This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2009 proceedings primary network deterministically To give a good protection, it is required that the interference to PUs on MAC layer should be limited by a certain target value We follow the similar definition of the interference constraint of packet error rate DP ER that primary network experiences on the channel as the expected packet collision rate of PUs with SU transmissions observed by the primary network [12] Assume the channel capacity is one and multiple PU packet collisions are negligible for SU transmissions The idle probμ and t ability of primary network is λ+μ T −τ (5) Since the duration of SU transmission in one frame longer than average idle time of primary network would create the drift, defined as the expected number of arrivals less the expected departures between state transmissions, leading to system instability, we assume (T − τ ) is less than average idle time λ1 , i.e (6) 0