Spectrum Sensing for Dynamic Spectrum Access of TV Bands (Invited Paper)

Transcription

1 Spectrum Sensing for Dynamic Spectrum Access of TV Bands (Invited Paper) Carlos Cordeiro, Monisha Ghosh, Dave Cavalcanti, and Kiran Challapali Wireless Communication and Networking Department Philips Research North America Briarcliff Manor, NY {Carlos.Cordeiro, Monisha.Ghosh, Dave.Cavalcanti, Abstract In this paper we address the issue of spectrum sensing in cognitive radio based wireless networks. Spectrum sensing is the key enabler for dynamic spectrum access as it can allow secondary networks to reuse spectrum without causing harmful interference to primary users. Here we propose a set of integrated medium access control (MAC) and physical layer (PHY) spectrum sensing techniques that provide reliable access to television (TV) bands. At the MAC level, we propose a two-stage spectrum sensing that guarantees timely detection of incumbents while meeting the quality of service (QoS) requirements of secondary users. At the PHY level, we introduce FFT-based pilot energy and location detection schemes that can detect a TV signal on a TV channel at levels as low as -116 dbm. We have evaluated these schemes through simulation and prototyping and show their effectiveness, reliability, and efficiency. These mechanisms are also part of the current IEEE draft standard which is based on cognitive radio technology. Keywords cognitive radio, dynamic spectrum access, spectrum sensing, MAC, protocol I. INTRODUCTION Most of the radio frequency spectrum is allocated, although much of it is unused. The enormous growth in the wireless industry has come from using only a small part of the wireless spectrum, nominally less than 10% under 3 GHz. There is growing evidence of scarcity and overcrowding in these bands reflected, for example, by the price paid for 3G cellular spectrum ( 90 billion in Europe). However, measurements have shown that other parts of the spectrum - although allocated - are virtually unused, and known widely as spectrum white spaces. These white spaces vary from place to place and time to time. With recent developments in Cognitive Radio (CR) [1][2][3] techniques, it becomes possible to harness these spectrum white spaces without causing harmful interference to incumbent users. Such dynamic spectrum sharing enabled by CRs, coupled with spectrum regulation to provide tiered access rights, is expected to dramatically increase the utilization of spectrum and enable a whole new class of applications. Tiered access rights may be described as incumbent users endowed with pre-emptive access to their spectrum, whereas the secondary CRs have access rights for opportunistic usage in spectrum white spaces on a non-interfering basis. A key building block in any CR system is the spectrum sensing function, which consists of (a) sensing algorithms to quickly and robustly detect the presence of incumbent signals, and (b) well-designed coordination and communication protocols. Since incumbent users must be detected at a power level of an order of magnitude (or more) below thermal noise floor, long sensing (also known as integration or quiet) times may be required thus precluding the support of real-time applications by the wireless network. In this paper we focus on two breakthrough spectrum sensing technologies based on pilot detection that provides quick and robust sensing, and that together with novel MAC schemes proposed, provide a means for the wireless network to support QoS sensitive applications without disruptions while protecting incumbents from harmful interference. The rest of this paper is organized as follows. In Section II we provide an overview of spectrum sensing requirements in the context of the CR-based IEEE standard [6]. In Section III we delve into the various specifics about spectrum sensing with respect to signal processing, baseband and MAC layer support. We then propose spectrum sensing algorithms and associated MAC protocols that jointly enable an effective and cohesive spectrum sensing function. The performance evaluation of these schemes is then given in Section IV. Finally, Section V concludes this paper. II. SPECTRUM SENSING REQUIREMENTS Spectrum sensing is the process of periodically and dynamically monitoring a given radio spectrum band (e.g., VHF and UHF TV bands) in order to determine its availability for use on a non-interfering basis. Spectrum sensing is a mandatory functionality in any CR-based wireless system that shares spectrum bands with primary services, such as the IEEE standard [5][6], which proposes to reuse vacant spectrum in the TV broadcast bands. Given its more advanced stage of development, in this section we describe the main requirements for spectrum sensing in the IEEE standard. Throughout the rest of the paper, we also use IEEE as an example application of the proposed spectrum

2 sensing mechanisms, although their underlying ideas could also be applicable to other wireless systems. A. Incumbents to Protect The spectrum sensing mechanisms in are designed to protect to two types of incumbents, namely, the TV service and wireless microphones. In particular, wireless microphones are licensed secondary users of the spectrum, and are allowed by FCC to operate on vacant TV channels on a non-interfering basis (please refer to Part 74 of the FCC rules) 1. Contrary to detection of TV transmission which takes place over 6 MHz channels (in the USA) with high transmit power, detection of wireless microphone operation is much harder as they transmit at a much lower power (typically 50 mw for a 100 m coverage range) and occupy much smaller bandwidths (200 KHz). B. DFS Timing Requirements The Dynamic Frequency Selection (DFS) timing parameters define the requirements that the secondary wireless system must adhere to in order to effectively protect the incumbents. Table 1 illustrates the key DFS parameters defined within , and which are based on the DFS model ordered by the FCC for the 5 GHz band [8]. Four key parameters are the (CDT), the Incumbent Detection Threshold (IDT), the Probability of Detection (PD) and the Probability of False Alarms (PFA). The CDT defines the time during which an incumbent operation can withstand interference before the system detects it. It dictates how quickly an system must be able to detect an incumbent signal exceeding the IDT with a probability of detection greater or equal to PD and probability of false alarm lower than or equal to PFA. Once the incumbent signal is detected at a signal level higher than IDT, two other new parameters have to be considered, namely, Channel Move Time (CMT) and Channel Closing Transmission Time (CCTT). The CCTT is the aggregate duration of transmissions by devices during the CMT. Parameter TABLE 1 SELECTED DFS PARAMETERS Value for Wireless Microphones Value for TV Broadcasting 2 sec 2 sec Channel Move Time (Inservice monitoring) Channel Closing Transmission Time (Aggregate transmission time) Incumbent Detection Threshold 2 sec 2 sec 100 msec 100 msec -107 dbm (over 200KHz) -116 dbm (over 6MHz) Probability of Detection 90% 90% Probability of False Alarm 10% 10% The DFS parameters are critical to the design of efficient spectrum sensing mechanisms that offer the required protection to the incumbents, while causing minimal impact on the operation of the network. III. PROPOSED SPECTRUM SENSING MECHANISMS The topic of spectrum sensing algorithms for detection of incumbent signals has recently been receiving a lot of attention [6][17][18]. Within IEEE , a number of techniques such as energy detection (full bandwidth and pilot), ATSC field sync detection, cyclostationary detection, spectral correlation, multi-resolution spectrum sensing and analog auto correlation have been proposed and evaluated via simulations using captured real-world ATSC signals. In this paper, we propose a novel method based on detecting the pilot energy and location that can be used with either one or multiple sensing dwells (windows 2 ), and hence fits well with the MAC sensing architecture by allowing the QoS of secondary services to be preserved despite the regularly scheduled sensing windows. At the MAC layer, existing work has not addressed the impact of spectrum sensing on QoS support to secondary users [12][13][14]. The IEEE h [4] standard includes basic mechanisms to quiet channels, but does not deal with the protocol mechanisms to synchronize quiet periods of overlapping networks or to guarantee seamless operation in the presence of incumbents. The two stage sensing, incumbent detection and notification, and synchronization schemes described in this section, however, have been designed for operation in highly dynamic and dense networks and have been adopted in the current draft of the IEEE standard [6]. Different variations of these schemes are also used in the Cognitive MAC (C-MAC) protocol introduced in [15]. A. Spectrum Sensing Algorithms In keeping with the general rule of IEEE 802, the draft standard cannot specify receiver algorithms. However, is a special case in which spectrum sensing is a very important feature of the standard even though its actual implementation will be in the receivers. Hence, the objective of the group is to define requirements for sensing that have to be met by all manufacturers. This specification of sensing requirements is ongoing [9]. As presented in Section II, the principal metrics for characterizing a sensing algorithm are PD and PFA. In our discussion throughout the rest of this paper, we use Probability of Missed Detection (PMD) instead of PD. The PMD is the probability of failing in detecting an incumbent user which is present on the sensed channel, and is defined as PMD = 1 - PD. Both PMD and PFA are functions of the received SNR and threshold. Ideally, one would like to have PMD = 0.0 and PFA = 0.0. However, in a practical situation these will be hard to achieve and it might be more 1 Throughout this paper, the terms incumbent and primary services are used interchangeably. 2 Throughout this paper, the terms sensing windows and dwells are used interchangeably.

3 reasonable to allow acceptable ranges for PFA of and PMD of From the incumbent protection point of view, a higher PFA is more tolerable than a higher PMD. There are two main approaches to spectrum sensing: energy detection and feature detection. Energy detection is used to determine presence of signal energy in the band of interest, and is followed by feature detection to determine if indeed the signal energy is due to the presence of an incumbent. Since will be implemented in the TV bands, the digital incumbent signals could be either ATSC (North America), DVB-T (Europe), or ISDB (Japan). In this paper we consider feature detection only for ATSC. The ATSC signal has a number of features that could be exploited for feature detection algorithms: (a) PN 511 sequence: The ATSC signal has a 511-symbol long PN sequence that is inserted in the data stream every 24.2 ms. Since this is quite infrequent, averaging over more than one field would be necessary for detection, leading to longer detection times. (b) Pilot: The ATSC signal uses a 8-VSB (Vestigial Side Band) modulation with signal levels (-7,-5,-3,- 1,1,3,5,7). A DC offset of 1.25 is added to this at baseband to effectively create a small pilot signal to enable carrier recovery at the receiver. (c) Segment-synch: the ATSC data is sent in segments of 828 symbols. At the beginning of each segment a 4- symbol sequence (5,-5,-5,5) is sent. Detection of this sequence can be used in a feature detector. (d) Cyclostationarity: Since the ATSC signal is a digital signal with a symbol rate of MHz, cyclostationary detectors may be used as a feature detector. The main problem with any feature detection method for ATSC is the requirement of detection at a very low signal level (-116 dbm see IDT in Table 1). Most of the synchronization schemes designed for ATSC receivers fail at these low signal levels and the detector may require large number of samples to average over for a reliable detection. In Section IV.A we present simulation results of a feature detection algorithm based on detecting the ATSC pilot. The ATSC VSB signal has a pilot at the lower band-edge in a known location relative to the signal. For this description, we will assume that the signal to be sensed is a band-pass signal at a low-if of 5.38 MHz with the nominal pilot location at 2.69 MHz and is sampled at MHz. However, the basic steps can be implemented with suitable modifications for any IF and sampling rate. The essential features of the proposed method are as follows: (1) Demodulate the signal to baseband by the nominal frequency offset of f c = 2.69 MHz. Hence, if x(t) is j2 f ct the real, band-pass signal at low-if, y( t) = x( t) e is the complex demodulated signal at baseband. (2) Filter y(t) with a low-pass filter of bandwidth, e.g., 40kHz (+/-20kHZ). The filter bandwidth should be large enough to accommodate any unknown frequency offsets. (3) Down-sample the filtered signal from MHz to 53.8 KHz, to form the signal z(t). (4) Take FFT of the down-sampled signal z(t). Depending on the sensing period, the length of the FFT will vary. For example, a 1 ms sensing window will allow a 32- point FFT while a 5 ms window will allow a 256- point FFT. (5) Determine the maximum value, and location, of the FFT output squared. Signal detection can then be done either by setting a threshold on the maximum value, or by observing the location of the peak over successive intervals. Instead of the FFT, other well-known spectrum estimation methods, such as the Welch periodogram can also be used in step (4) above. The basic method described above can be adapted to a variety of scenarios as described below: (1) Multiple fine sensing windows, e.g. 5 ms sensing dwells every 10 ms. The 256-point FFT outputs squared from each sensing window can be averaged to form a composite statistic as well as the location information from each measurement can be used to derive a detection metric. (2) If a single long sensing window, e.g. 10 ms is available, a 512-point FFT or periodogram can be used to obtain better detection performance. The parameters of the sensor can be chosen depending on the desired sensing time, complexity, probability of missed detection and probability of false alarm. Detection based on location is robust against noise uncertainty since the position of the pilot can be pinpointed with accuracy even if the amplitude is low due to fading. Various combing schemes can be developed for both pilot-energy and pilot-location sensing. (1) Pilot-energy sensing: For a single sensing window, the FFT output is simply squared and the maximum value is compared to a threshold. For multiple sensing dwells, there are 2 possibilities: (i) the decision from each dwell is saved and a hard-decision rule is applied to declare signal detect if the number of positives is greater than a certain number, or (ii) the square of the FFT output of all dwells is averaged and the maximum level is compared to a threshold. The choice of threshold in all cases is determined by the desired PFA. (2) Pilot-location sensing: This is usually used for multiple dwells. The location of the maximum value of the FFT output squared is compared between multiple dwells. If the distance is less than a prescribed threshold, the signal is declared detected. Another method is to count the number of times a particular frequency bin is chosen as the location of the maximum: if greater than a certain threshold, the signal is declared detected. B. Spectrum Sensing at the MAC In order to maximize the reliability and efficiency of the spectrum sensing algorithms described in the previous

4 subsection, and meet the CDT requirement for detecting the presence of incumbents, the network can schedule networkwide quiet periods for sensing. During these quiet periods, all network traffic is suspended and stations can perform sensing more reliably. In , for example, the base station (BS) is responsible for managing and scheduling these quiet periods. To meet these requirements while satisfying the QoS requirements of the secondary network, we propose a two stage sensing (TSS) management mechanism. The TSS mechanism enables the network to dynamically adjust the duration and the frequency of quiet periods according to the sensing windows required by the sensing algorithms (see Section III.A) in order to protect the incumbents. In the first stage, multiple short quiet periods are scheduled to attempt to assess the state of the sensed radio spectrum without causing impact to the secondary network performance. In the second stage, more time consuming quiet periods can be scheduled in case the target spectrum needs to be sensed for a longer period of time. In addition to the TSS which provides timely detection mechanism, we also introduce a notification mechanism through which devices (e.g., the consumer premise equipments (CPEs) in IEEE terminology) can report the results of the sensing process back to the BS (see Section III.B.2). To ensure the effective use of quiet periods to improve sensing reliability, nearby networks must also synchronize their quiet periods. In Section III.B.3, we discuss how coexisting cells can synchronize their quiet periods. The TSS, notification, and synchronization mechanisms proposed here have been incorporated into the current draft MAC standard. In the next subsection we present these schemes within the context of , and evaluate their performance in Section IV.B. B.1 Quiet Periods Management and Scheduling The TSS quiet period management mechanism has different time scales, namely, a short (or fast) sensing period that can be scheduled regularly with minimal impact on the users QoS, and a long (or fine) sensing period that can be used to detect a specific type of incumbent signal. The short and long sensing periods correspond to the first and second stage of the TSS, respectively. The TSS presented here is a more general and enhanced version of the MAC sensing scheme introduced in [15][16]. Within IEEE , the first stage of TSS is termed as intra-frame sensing, while the second stage is called interframe sensing. Intra-frame Sensing: This stage uses short quiet periods of less than one frame size 3. The MAC allows only one intra-frame quiet period per frame and it must be scheduled always at the end of the frame. This is important to ensure nearby cells can synchronize their quiet periods, as discussed in Section III.B.3. Based on the results of spectrum 3 The MAC is based on a periodic superframe structure. A superframe contains 16 frames of 10 msec each for a total duration of 160 msec. sensing done over a number of intra-frame quiet periods, the BS decides whether to schedule an interframe quiet period over multiple frames in order to perform more detailed sensing. Inter-frame Sensing: This stage is defined as taking longer than one frame size and it is used when the sensing algorithm requires longer sensing durations. Since a long quiet period may degrade the performance for QoS sensitive traffic, the allocation and the duration of the inter-frame sensing stage should be dynamically adjustable by the BS in a way to minimize the impact on the users QoS. The TSS mechanism in IEEE is illustrated in Figure 1. A first stage involving several intra-frame sensing periods can be followed by a longer inter-frame sensing period, if needed to detect the specific signature of a signal detected during the first stage. Considering the fact that incumbents in TV bands do not come on the air frequently, only the intraframe sensing stage will be used most of the times, and so QoS is not compromised. The longer inter-frame sensing stage will step in only when required. In , the BS broadcasts the schedule and the durations of the intra-frame and inter-frame quiet periods in the superframe control header (SCH), which is transmitted at the beginning of every superframe. This method incurs minimal overhead and allows the scheduling of quiet periods well in advance, which enables a tight synchronization of quiet periods amongst neighboring systems. The BS can also schedule quiet periods on an on-demand basis using management frames specified in the draft [7]. One of the major benefits of the TSS mechanism is allowing the CR network to meet the stringent QoS requirements of real time applications such as voice over IP, while ensuring the required protection to the incumbents. Figure 1 The TSS mechanism B.2 Incumbent Detection and Notification Once an incumbent is detected on an operating channel, say channel N, or in an adjacent channel (e.g. N+1 and N-1), the secondary system must vacate the channel, while satisfying the DFS requirements (CMT and CCTT) described in Table 1. In a cell-based system, such as , detection of an incumbent must be notified in a timely fashion to the BS, so it can take proper action to protect the incumbents 4. A number of 4 In this is referred to as incumbent detection recovery and is performed through the Incumbent Detection Recovery Protocol (IDRP). IDRP maintains a priority list of backup channels that can be used to quickly re-establish communication in the event of an incumbent appearance.

5 mechanisms are described in the draft standard to deal with these situations. For example, a CPE may notify the BS by using the UCS (Urgent Coexistence Situation) slots available within the MAC frame. Since the allocation of the UCS window is known to all CPEs, it can be used even when CPEs are under interference. As far as access method goes, both contention-based and CDMA can be used during the UCS window. Alternatively, the BS can poll CPEs to obtain feedback. In this case, the polled CPE can send a notification back to the BS, or else if no response is received from CPEs, the BS can take further actions to assess the situation such as scheduling additional quiet periods or even immediately switching channels. CBP packet was received/transmitted, respectively. By this mechanism, it has been shown [10] that co-channel networks are able to synchronize their quiet periods resulting in the arrangement depicted in Error! Reference source not found.. This way, sensing can be made with high reliability. BS 1 BS 2 Intra-frame sensing Inter-frame sensing Intra-frame sensing Inter-frame sensing Intra-frame sensing Intra-frame sensing Inter-frame sensing Inter-frame sensing Chanel Detection Time Intra-frame sensing Intra-frame sensing Inter-frame sensing Inter-frame sensing B.3 Synchronization of Quiet Periods BS 3 Time Self-coexistence amongst multiple overlapping CR networks is a key feature not only to efficiently share the available spectrum, but also to ensure the required protection to the incumbents. For instance, multiple secondary networks may operate in the same geographical region, and in case nearby networks share the same channel, it is paramount that they are able to synchronize their quiet periods, since transmissions during sensing windows could increase PFA considerably. For the case of the standard, it provides a comprehensive coexistence framework to enable overlapping networks to exchange information in order to share the spectrum and also synchronize their quiet periods. At the core of this framework is the Coexistence Beacon Protocol (CBP), which is based on the transmission of CBP frames (beacons or packets) by CPE and/or BSs. The CBP packets are transmitted during the coexistence windows that can be open by the BS at the end of a frame. During these windows, CPEs in overlapping areas can send CBP packets using a contentionbased access mechanism. These packets may be received by neighboring BSs or by CPEs in neighboring cells, which forward them to their corresponding BSs. The CBP packets carry information needed for establishing time synchronization amongst neighboring cells, as well as the schedule of the quiet periods. For the purpose of synchronization, CBP packets carry relative timestamp information about their networks. Mathematically speaking, when BS i, responsible for network i, receives a CBP packet from network j, controlled by BS j, it shall adjust the start time of its superframe if, and only if, the following convergence rule is satisfied: ( Frame _ Numberj Frame _ Numberi ) FDC + FS FDC + GuardBand SymbolSize Transmission _ Offset Reception _ Offset 2 Where Frame_Number is the frame number within the superframe, FDC is the frame duration code (equal to 10ms), FS is the number of frames per superframe (equal to 16), GuardBand is a few OFDM symbols long to account for propagation delays, SymbolSize is the size of an OFDM symbol, and Reception_Offset and Transmission_Offset are the index of the symbol number within the frame where the Intra-frame sensing Inter-frame sensing Figure 2 Synchronization of quiet periods IV. PERFORMANCE EVALUATION Transmission We have evaluated the proposed spectrum sensing algorithms and MAC layer mechanism for supporting spectrum sensing via simulations. The MAC has been implemented in the OPNET network simulator. For the spectrum sensing algorithms, we have a simulation platform as well as have built a real-time prototypical sensor that can detect digital TV signals at a low signal level of 116dBm, as per required (see Table 1). A. Spectrum Sensing Algorithms Simulations were performed using the sensing algorithm described in Section III with the 12 signals specified by the IEEE committee [19]. These signals cover the range of real-world signal conditions with respect to frequency offsets and multipath fading and hence are a good test for any sensing algorithm. The sensing time used was a multiple of 5 ms dwells, which allowed the use of a 256-point FFT. Figure 3 shows the PMD versus SNR curve for a PFA of 5%, for a single 5 ms sensing window, using the pilot-energy sensing method described earlier. The results for all 12 signals, as well as the average over these signals are shown. It is clear that there is a wide variation in the performance of the pilotenergy detector and this is because of multipath fading attenuates the pilot on some of these signals. Nevertheless, an average PMD = 0.1 can be achieved at SNR = -18 db, which is 33 db below the threshold of visibility of 15 db for a DTV signal in AWGN. Figure 4 shows the performance of the pilot-energy detector when 6 sensing dwells of 5 ms each are used, for a PFA = 4%. With this increased sensing time, the average performance improves so that PMD = 0.01 at SNR = -20 db.

6 Figure 3 - Performance of pilot-energy detection with 5 ms sensing time Figure 5 shows the performance of the pilot-location detector with PFA = 3%. This detector works as follows: the location of the maximum of the FFT-output squared averaged over the first 3 dwells is compared to the location of the maximum of the FFT-output squared over the next 3 dwells. If these 2 locations are within one frequency-bin apart, the sensor declares a detection. This method gives a performance of PMD = 0.01 at SNR = -19 db which is about 1 db worse than the pilot-energy detector. However, the threshold for the pilot-location detector is completely independent of the noise-level and hence is robust to uncertainties in the measurement of noise at the sensing receiver. Figure 6 shows the sensing performance of the pilotenergy detector and Figure 7 shows the performance of the pilot-location detector, with a 2 db noise uncertainty, i.e. the noise variance was assumed to have an uniform distribution over [-2 db, 2 db]. The pilot-energy detector has an increased PFA of 0.21, whereas the performance of the pilot-location detector is essentially unchanged. Figure 4 - Performance of pilot-energy detection with 5 ms X 6 sensing window Figure 6 - Performance of pilot-energy detector with 2 db noise uncertainty B. Spectrum Sensing at the MAC In this section we evaluate the MAC layer mechanisms for supporting spectrum sensing. For all the MAC simulations, the OFDM symbol size is fixed at 310µsec, each symbol has 1536 data carriers, and the modulation/coding used is 64-QAM rate 2/3, providing a capacity of about 19.8Mbps per 6MHz TV channel. The MAC frame size is 10ms, the superframe contains 16 MAC frames, and the packet size is 1Kbyte. B.1 Impact of the TSS Mechanism on Performance Figure 5 - Performance of pilot-location detection for 5 ms x 6 sensing time To evaluate the impact of TSS on the overall system efficiency, we consider 1 BS and a total of 127 CPEs. We fix the duration of each intra-frame sensing to 1ms, and program the BS to allocate an intra-frame sensing window every 4 MAC frames. Considering that CDT is 2sec and the MAC frame size is 10ms, this would result in approximately 200

7 frames per CDT. In turn, this results in about 50 MAC frames carrying an intra-frame sensing window per CDT, or around 50ms of intra-frame quiet periods for sensing per CDT. Also, note that this is done for each channel in use by the secondary network. transmitting a channel switch command to all nodes. Figure 10 shows the instantaneous throughput curves for nodes 5 and 6, and how this is affected by periodic sensing (dips in the curve). Figure 8 Efficiency of the TSS scheme Figure 7 - Performance of pilot-location detector with 2 db noise uncertainty Radio range of TV station Figure 8 shows the efficiency of the protocol with and without the TSS scheme and for different number of channels 5. Here, efficiency is defined as the fraction of the channel capacity that is used for the actual payload transmission (i.e., capacity made available at the MAC service access point). For these simulations, each channel is carries constant bit rate traffic where the upstream aggregate traffic is fixed to 3Mbps while the downstream traffic is varied from 2Mbps to 22Mbps. As we can see, for low to medium loads the TSS has nearly no impact on performance. Only at high loads will the quiet periods impact the capacity. Even so, the overall impact is not sizeable. Node 5 BS on TV channel A Node 6 TV station on channel A B.2 Network Resilience Here we study the capability of the MAC in guaranteeing protection to incumbent services while seamlessly resuming communication in vacant channels. This includes the joint application of the incumbent detection (sensing), notification and recovery (IDRP) mechanisms. For this analysis, we consider the scenario depicted in Figure 9 consisting of 1 BS and multiple nodes (i.e., CPEs). For this simulation, we impose that both stages of the TSS scheme are always executed. At a random point in time during the simulation run, a TV station begins operating co-channel with the secondary network and interferes with the nodes as shown in Figure 9. Through the TSS and polling mechanisms, the BS realizes (within CDT) that the nodes are not communicating and then initiates the recovery procedure by Figure 9 Simulation scenario As we can see from the highlighted portion of this figure, which indicates the appearance and detection of the of the TV signal, node 5 s throughput is unaffected since it is able to correctly receive the channel switch command from the BS. Node 6, however, does not receive the command from the BS, but is nevertheless able to restore normal communication through the IDRP protocol, as it knows the backup channel (in this case, channel B) used by the BS. In this simulation, the backup channel B is in fact a set of three contiguous bonded TV channels, which explains the significant increase in instantaneous throughput. The entire notification and recovery procedure for the whole network takes at most two MAC frames, which is much less than required by CMT and CCTT. 5 In this set of simulations, we considered that up to 3 channels can be used by BS simultaneously [20].

8 digital TV signals at a low signal level of 116dBm, and hence reliable satisfy the sensing requirements for reliable operation in TV bands. TV detection, notification and recovery Figure 10 Impact of incumbent appearance in network resilience B.3 Synchronization of Quiet Periods Synchronization of quiet periods is critical to the reliable and efficient spectrum sensing in the presence of multiple secondary networks. We have conducted extensive simulations to study how quickly and reliably overlapping networks converge to the same superframe start time, and therefore can easily align their quiet periods by using the information received through the CBP packets. The results are shown in Figure 11. The simulation consists of a number of networks (xaxis), placed randomly in square area (50x50, 100x100, and 150x150 km), with random start times and fixed range of 25 km. Each network periodically issues CBP packets as described in Section III.B.3, and synchronizes according to the convergence rule. On the y-axis is the convergence time in units of superframe. As can be seen, even with a large number of networks the convergence is very quick, even though the synchronization operation is completely distributed in nature. V. CONCLUSIONS This paper proposes PHY and MAC layer sensing techniques that enable efficient and reliable access to TV channels. By combining the proposed two-stage sensing mechanism at the MAC, with the proposed FFT-based sensing method to detect the pilot energy and location at the PHY, a secondary CR network can reliably detect a TV signal at very low signal levels. It has been shown that the pilot location detection is more robust to uncertainties in the measurement of noise at the sensing receiver than the pilot-energy detection. The MAC simulations indicate that the two-state sensing mechanism has nearly no impact on the cell s performance at low or medium loads, and only a small impact on the cell s performance at high loads. Furthermore, the simulation results also show that synchronization of the quiet periods amongst cells can be achieved in a timely manner. Although not described in this paper, we have built a real-time prototypical sensor using our proposed sensing method that can detect Figure 11 Convergence time in the synchronization of quiet periods [16] REFERENCES [1] J. Mitola et al., Cognitive Radios: Making Software Radios more Personal, IEEE Personal Communications, vol. 6, no. 4, Aug [2] J. Mitola, Cognitive radio: An integrated agent architecture for software defined radio, PhD Dissertation, Royal Inst. Technol. (KTH), Stockholm, Sweden, [3] S. Haykin, Cognitive Radio: Brain-Empowered Wireless Communications, in IEEE JSAC, vol. 23, no. 2, Feb [4] IEEE h Standard for Spectrum and Transmit Power Management Extensions, pdf, [5] C. Cordeiro, K. Challapali, D. Birru, and S. Shankar, IEEE : The First Worldwide Wireless Standard based on Cognitive Radios, in IEEE DySPAN, Nov [6] IEEE Working Group on Wireless Regional Area Networks, [7] IEEE draft standard, IEEE P TM /D0.3 Draft Standard for Wireless Regional Area Networks, doc. no , May [8] Federal Communications Commission (FCC), Revision of Parts 2 and 15 of the Commissions Rules to Permit Unlicensed National Information Infrastructure (U-NII) Devices in the 5GHz Band, ET Docket no , Nov. 18, [9] G. Chouinard, D. Cabric and M. Ghosh, Sensing Thresholds, IEEE , doc. no , May [10] C. Cordeiro et al., A PHY/MAC Proposal for IEEE WRAN Systems, IEEE , doc. no , March [11] S.-Y. Chang et al., IEEE WRAN Merger Framework, IEEE , doc. no , March [12] S. Seidel and R. Breinig, Autonomous Dynamic Spectrum Access System Behavior and Performance, in IEEE DySPAN, Nov [13] L. Ma, X. Han, and C.-C. Shen, Dynamic Open Spectrum Sharing for Wireless Ad Hoc Networks, in IEEE DySPAN, Nov [14] Q. Zhao, L. Tong, and A. Swami, Decentralized Cognitive MAC for Dynamic Spectrum Access, in IEEE DySPAN, Nov [15] C. Cordeiro and K. Challapali, C-MAC: A Cognitive MAC Protocol for Multi-Channel Wireless Networks, in IEEE DySPAN, April [16] C. Cordeiro, K. Challapali, and M. Ghosh, Cognitive PHY and MAC Layer for Dynamic Spectrum Access and Sharing of TV Bands, in IEEE International Workshop on Technology and Policy for Accessing Spectrum (TAPAS), August 2006.

9 [17] A. Sahai, R. Tandra, S. Mubaraq Mishra, N. Hoven, Fundamental design trade-offs in cognitive radio systems, in IEEE International Workshop on Technology and Policy for Accessing Spectrum (TAPAS), August [18] S. J. Shellhammer, Sai Shankar N, R. Tandra and J. Tomcik, Performance of Power Detector Sensors of DTV Signals in IEEE WRANs, in IEEE International Workshop on Technology and Policy for Accessing Spectrum (TAPAS), August [19] S. Shellhammer, V. Tawil, G. Chouinard, M. Muterspaugh and M. Ghosh, Spectrum Sensing Simulation Model, IEEE /0028r10, August [20] K. Challapali, C. Cordeiro, and D. Birru, Evolution of Spectrum-Agile Cognitive Radios: First Wireless Internet Standard and Beyond, in 2nd ACM International Wireless Internet Conference (WICON), August 2006.