914 IEEE TRANSACTIONS ONELECTROMAGNETIC COMPATIBILITY, VOL. 49, NO. 4, NOVEMBER 2007 Channel Clustering and Probabilistic Channel Visiting Techniques for WLAN Interference Mitigation in Bluetooth Devices Qixiang Pang and Victor C. M. Leung, Fellow, IEEE Abstract—Since Bluetooth and wireless local area network (WLAN) technologies both operate at the 2.4-GHz industrial, sci- entific, and medical (ISM) band, the two types of devices may suffer from mutual interference and performance degradations. In this paper, we propose two new techniques, channel cluster- ing and probabilistic channel visiting, to effectively improve the existing coexistence and interference mitigation mechanisms. The channel clustering technique employs statistical pattern recogni- tion to classify the status of Bluetooth channels more accurately. The probabilistic channel visiting is used to more equitably allo- cate the channel resources between Bluetooth and WLAN devices. The effectiveness of these techniques is quantified by simulations. Results show that both techniques are beneficial in improving the performance of the existing mechanisms. Index Terms—Bluetooth, coexistence, interference mitigation, wireless local area network (WLAN). I. INTRODUCTION I EEE 802.11 wireless local area networks (WLANs) [1] and IEEE 802.15.1 Bluetooth (BT) wireless personal area net- works (WPANs) [2] have been widely deployed. Because both systems operate in the 2.4-GHz industrial, scientific, and med- ical (ISM) frequency band, mutual interference between them may result in severe performance degradations [3], and coexis- tence of these systems over a shared electromagnetic spectrum becomes an important issue. Recently, the interference issue has attracted many research and standardization activities [3]–[8]. For example, the IEEE 802.15.2 standard [3] includes eight coexistence and interfer- ence mitigation mechanisms that can be classified into two cat- egories: collaborative and noncollaborative. The collaborative mechanisms can be used when the WLAN and BT devices are integrated within the same physical unit and exchanging infor- mation between them is feasible. The noncollaborative mech- anisms are designed for application in BT devices when they are not collocated with the interfering WLAN devices within the same physical units. In practice, since many WLAN and BT devices are independently implemented and installed, the Manuscript received October 3, 2006; revised February 27, 2007 and May 24, 2007. This work was supported in part by the Canadian Natural Sciences and Engineering Research Council under Grant STPGP 257684-02 and in part by the OPNET University program. This paper was presented in part at the IEEE VTC-Spring, May 2006. The authors are with the Department of Electrical and Computer Engineering, The University of British Columbia, Vancouver, BC V6T 1Z4, Canada (e-mail: qixiangp@ece.ubc.ca; vleung@ece.ubc.ca). Digital Object Identifier 10.1109/TEMC.2007.908262 noncollaborative mechanisms are more practical. In the follow- ing discussions, we shall concentrate on the noncollaborative mechanisms. The existing noncollaborative mechanisms all use similar techniques to detect the presence of WLANs in the ISM band and classify the BT channels as “bad” if they are subject to WLAN interference, or “good” otherwise. For example, each BT device can estimate the packet-error rate (PER) or received signal strength indication (RSSI) for each channel, and a chan- nel is marked “bad” if the PER or RSSI value exceeds a prede- fined threshold [3]–[5]. The device then modifies its frequency- hopping pattern to avoid the “bad” channels, or chooses not to transmit when hopping onto a “bad” channel. The existing channel classification methods are simple, but they are effec- tive only when a single BT piconet exists in the area of inter- est. When there are multiple piconets within radio range, these methods can be misled as a high PER or RSSI can result when more than one piconet hops onto the same channel. To address this issue, sensing of actual WLAN carriers could be an effec- tive option, which is, however, quite complex and possibly too costly to implement in low-cost BT devices due to physical layer differences. Besides channel classification, a related problem is how of- ten a BT piconet should visit the channels after they have been classified. In the existing mechanisms, once a channel is marked as “bad,” it will not be visited at all. However, channel condi- tions are subject to changes and the channel classification can be misled by the presence of other BT piconets, as discussed earlier. When multiple WLANs and/or BT piconets coexist, the number of “good” channels can be very low and the BT de- vices may be starved of radio resources due to their excessive courtesy. Therefore, the existing coexistence mechanisms may not work effectively in some practical situations. This paper con- tributes novel solutions to the respective problems discussed ear- lier by proposing the channel clustering classification method and probabilistic channel visiting. The current work substan- tially improves and extends our preliminary work [9] in the following aspects: 1) a detailed analysis of the problems in the existing PER-based and RSSI-based channel classification methods; 2) a detailed analysis of the problem in the existing channel visiting methods; 3) a refined channel classification method compared with [9] and other previous work [3]–[7]; 4) an analytical evaluation of the proposed scheduling mecha- nism; and 5) additional simulation scenarios and more detailed discussions on the performance results. 0018-9375/$25.00 © 2007 IEEE