IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 2, NO. 2, APRIL 2013 227 Adaptive Spectrum Sensing for Time-Varying Channels in Cognitive Radios Hao He, Geoffrey Ye Li, and Shaoqian Li Abstract—In this article, we develop a novel adaptive spectrum sensing scheme to improve the throughput of cognitive radio (CR) users. The new scheme takes the variation of wireless channels into consideration and requires no priori knowledge of primary user activity statistics. At the beginning of each time frame, this adaptive sensing scheme adjusts the spectrum sensing parameters according to the latest sensing results and channel state information (CSI) of the time-varying channels. Numerical results show that the proposed adaptive spectrum sensing scheme can significantly outperform the traditional spectrum sensing scheme. Index Terms—Cognitive radio; spectrum sensing; time-varying channel. I. I NTRODUCTION D IFFERENT from traditional radios, cognitive radio (CR) is capable of sensing its environments and promptly reconfiguring its system parameters [1]. Therefore, how to provide more spectrum access opportunities to CR users while limiting interference to the primary users is an important aspect in CR technologies. The sensing duration can impact the transmission oppor- tunities of CR users in several ways. The longer sensing duration leads to more accurate sensing but also reduces the available time for data transmission. In recent years, the tradeoff between the spectrum sensing and transmission gets more and more attention [2-5]. The sensing duration and channel selection for periodic spectrum sensing have been investigated in [2]. In [3], a cooperative decision method for sensing parameters has been proposed, where the transmission duration is adaptively adjusted according to the number of CR users. An parametric adaptive spectrum sensing architecture is proposed in [4], which takes into account the statistics of the channel availability. In [5], an adaptive compressive spectrum sensing algorithm has been proposed, which can adaptively adjust compressed measurements without any sparsity estima- tion efforts. Note that these works all assume that channels are time-invariant. The case that CR users experience time-varying channels is considered in [6], where the channel availability is as- Manuscript received November 5, 2012. The associate editor coordinating the review of this letter and approving it for publication was L. Lampe. H. He is with the National Key Lab of Science and Technology on Communications, University of Electronic Science and Technology of China, Chengdu, China, and with the School of ECE, Georgia Institute of Technology, Atlanta, GA, USA (e-mail: tracyhehao@gmail.com). G. Y. Li is with the School of ECE, Georgia Institute of Technology, Atlanta, GA, USA. S. Li is with the National Key Lab of Science and Technology on Communications, University of Electronic Science and Technology of China, Chengdu, China. This work was supported in part by the NSF under Grant No. 1247545. Digital Object Identifier 10.1109/WCL.2013.012513.120810 Data ACK/ NAK Pilot Sensing s T f T P T CSI Fig. 1. Frame structure. sumed to change much slower than the sensing/transmission activities. In this article, we investigate adaptive spectrum sensing scheme for time-varying channels under more realistic assumptions. At the beginning of each time frame, the sensing strategy is decided by the previous sensing result and channel condition. Our adaptive spectrum sensing scheme aims at maximizing the throughput of CR users while maintaining protection for the primary users. The new adaptive sensing scheme requires no priori knowledge of primary user activity statistics. II. SYSTEM MODEL AND PROBLEM FORMULATION In this section, we illustrate the system model and formulate the optimization problem for the spectrum sensing. Most of the previous sensing schemes are based on the priori knowledge of the primary user activities. Due to huge difference among primary networks, this assumption is not satisfied sometimes. In this article, the only information known by CR user is that the busy/idle state transition of primary networks, which is assumed to occur at the beginning of each time frame. That means the primary network is either stay active or stay idle in a whole time frame. The frame structure of CR user is illustrated in Fig.1, where the length of each frame is T f . We consider the case that the CR user always has data packets to transmit. At the beginning of each time frame, CR user senses the availability of spectrum within a period of T s . Once the spectrum sensing result indi- cates the channel is idle, CR user transmits pilot to CR receiver and obtain channel state information (CSI) within a period of T p . The CSI is fed back through a dedicated error-free feedback channel without delay. After data are transmitted, the receiver acknowledges every successful or unsuccessful transmission by error-free ACK or NAK, respectively. For CR users, we use a block fading model to characterize the time-varying channel, i.e., the channel state keeps constant during each frame. It is well-known that block fading channel can be modeled as an ergodic first order finite-state markov channel (FSMC) [7]. For a Rayleigh fading channel with additive white Gaussian noise (AWGN), the received signal- to-noise ratio (SNR) is a random variable with probability 2162-2337/13$31.00 c  2013 IEEE