Multicarrier communication techniques for spectrum sensing and communication in cognitive radios

IEEE Communications Magazine • April 200880 0163-6804/08/$25.00 © 2008 IEEE INTRODUCTION The demand for ubiquitous wireless services has been on the rise in the past and is expected to remain the same in the future. Unfortunately, the vast majority of available spectral resources have already been licensed. It thus appears that there is little or no room to add any new services unless some of the existing licenses are discon- tinued. On the other hand, studies have shown that vast portions of the licensed spectra are rarely used. This has initiated the idea of cogni- tive radio (CR), where secondary (i.e., unli- censed) users are allowed to transmit and receive data over portions of spectra when primary (i.e., licensed) users are inactive, demanding that the secondary users (SUs) be invisible to the primary users (PUs). To fulfill the invisibility requirement, SUs need to sense the spectrum, and this involves some sort of spectral analysis. When the nature of the licensed signal is known, such analysis can be performed through feature (e.g., pilot) detec- tion. This is the case in IEEE 802.22, which is currently being designed to operate in the TV bands. However, in the general case where such knowledge is unavailable, spectral analysis has to rely on energy detection, with likely higher requirements for sensitivity and frequency reso- lution. MOTIVATION Recently, multicarrier methods have been recog- nized as potential candidates for the physical layer of CR systems. By assigning SUs to the subcarriers that coincide with the portions of the spectrum not used by PUs, multicarrier methods provide much flexibility to fill in the spectral holes and thus to best harness the available resources. Moreover, orthogonal frequency-divi- sion multiplexing (OFDM), the most popular multicarrier method, has been introduced as the first candidate for this purpose [1]. In addition, it has been noted that the fast Fourier transform (FFT) as part of the OFDM demodulator can be used for spectral analysis, to identify the pres- ence/absence of the active PUs. However, a number of shortcomings of OFDM in the appli- cation of CR have been noted as well. The short- comings originate from the large side-lobes of the frequency response of the filters that charac- terize the subcarrier channels. These side-lobes, in turn, result in significant interference among subcarriers that originate from different SUs, and between PUs and SUs. Interestingly, as of spring 2006, the working document of the IEEE 802.22 standard covered both frequency-division duplex (FDD) and time-division duplex (TDD) operation. In recent meetings of the standardiza- tion group, the standard has been modified to include TDD operation only. While this decision is likely inspired by the IEEE 802.16 MAC layer, it also suggests that FDD operation has been ABSTRACT In this tutorial article we review different multicarrier communication methods for the physical layer of cognitive radio systems. There, secondary users need to dynamically and reliably determine spectral holes, and transmit data in these resources without interfering with other parts of the frequency band. To satisfy the first, each SU has to be equipped with a spectrum analyzer. To satisfy the second, it is widely accepted that a multicarrier modulation tech- nique should be adopted. Moreover, to maxi- mize efficiency, it has been recognized that the side-lobes of each subcarrier band must be mini- mized. Much of the attention in the present lit- erature emphasizes on the use of conventional OFDM, exploiting the fact that fast Fourier transform (FFT) as part of the OFDM modula- tor can also be used for channel sensing. Herein, we discuss the performance of OFDM, and also introduce filterbanks for multicarrier communi- cation and spectral analysis in a CR setting. Moreover, the multitaper method has been pro- posed as an effective method for spectrum anal- ysis. Our article provides an insight into the pros and cons of these technologies. COGNITIVE RADIO COMMUNICATIONS AND NETWORKS Behrouz Farhang-Boroujeny and Roland Kempter, University of Utah Multicarrier Communication Techniques for Spectrum Sensing and Communication in Cognitive Radios FARHANG LAYOUT 3/24/08 5:13 PM Page 80 IEEE Communications Magazine • April 2008 81 suspended because of the leakage issues of OFDM. In fact, Philips and France Telecom as the group members proposed the use of a filter- bank multicarrier technique, termed offset quadrature amplitude modulation (OQAM) [2], to ease the leakage problem. Returning to the issue of spectrum sensing, in order to reliably detect the available spectrum holes, the channel sensing mechanism needs to feature a high spectral dynamic range (SDR). If this is not the case, SUs may either interfere with low-power PUs or not be able to best har- ness the wireless resources. Furthermore, in order to increase bandwidth efficiency, receivers need to have acceptable out-of-band rejection capabilities. Unfortunately, as we outline, FFT as part of an OFDM data transmission system is neither able to provide a sufficiently high SDR for channel sensing, nor can it suit the FCC’s envisioned out-of-band rejection requirements. Other candidates for multicarrier communi- cations in CR networks are filterbank-based sys- tems [3]. While filterbank multicarrier communication seems to be less familiar to the cognitive radio community, it has received con- siderable attention from researchers in both the communications and signal processing areas. Over the past three decades, three classes of such systems have been introduced. Saltzberg [4] was the first to propose a filterbank multicarrier communication system using a special QAM technique. Prior research that initiated this development was performed by Chang [5]. Developments in digital subscriber line (DSL) technologies led to two other classes of filter- bank multicarrier communication systems: fil- tered multitone (FMT) [6] and discrete wavelet multitone (DWMT) modulation [7]. Of these, DWMT has been further developed recently and renamed cosine-modulated multitone (CMT) [8]. A short overview of these methods is deferred to a later section. We note that the shortcomings of OFDM/ FFT in terms of leakage and reduced sensitivity may not matter much in today’s CR systems. For example, the FCC requires IEEE 802.22 to maintain large guard-bands to adjacent TV channels. Also, the standard does not include frequency-division multiple access (FDMA) operation, where, out of the entirety of the CR “band,” different clusters of subcarriers may be assigned to different users. To elaborate, the problem of the large side lobes of OFDM may only affects the subcarriers at the band edges between the users. When channel access is facili- tated in a TDD fashion, where the entire spec- trum is assigned to only one user at a time, the overhead introduced through guard bands at the band edges is negligible and thus acceptable. As such, the adoption of OFDM to CR is very rea- sonable. Moreover, OFDM technology is very well understood and very mature, and chip sets are readily available at low cost. However, as soon as a true multi-user multicarrier FDMA/ FDD operation is adopted, the limitations of OFDM/FFT mentioned above are likely to turn out to be significant, and a filterbank-based solu- tion may be preferred. Compared to OFDM, however, filterbank multicarrier is less well understood and studied. Hence, its deployment requires significant research in the future. It is our intention to motivate such research and to investigate the pros and cons of OFDM and fil- terbank multicarrier in the CR setting. In the following, we show that while OFDM appears as the natural candidate for multicarrier communications, it leads to inefficient spectrum utilization because of severe spectral leakage. Opposed to this, filterbank multicarrier can overcome the spectral leakage problems of OFDM and therefore lead to higher spectral efficiency. Regarding spectral sensing, we demonstrate that the FFT as part of the OFDM demodulator may provide insufficient SDR, and thus, CRs may be unable to detect low-power users. As the channel sensing tool, we present a novel filterbank-based spectrum analyzer as an alternative to the near-optimum Thomson’s mul- titaper method (MTM), [9, 10]. We demonstrate that compared to the MTM, the filterbank spec- trum analyzer achieves almost identical perfor- mance at much reduced complexity. In addition, when filterbanks are already used for data com- munication, filterbank channel sensing comes at virtually no additional computational cost. THE NOISE FLOOR IN CR NETWORKS The concept of the noise floor is a means for evaluating the effective background noise in heavily utilized parts of the spectrum. Along these lines, interference from SU sources may raise the noise floor for PUs. To quantify the effect of such sources on the PUs, the FCC spec- trum policy task force has recommended the interference temperature as a new performance metric [9]. Essentially, if different users’ chan- nels are at some proximity, spectral leakage among the users is likely. This will increase the background noise, which, in turn, is equivalent to operating the system in an environment with a higher temperature. Traditionally, in legacy systems, by imposing frequency masks as part of the system requirement as well as demanding sufficient separation among different user chan- nels, leakage effects and therefore the interfer- ence temperature are minimized. Mathematically, the interference temperature in a desired channel of width ∆f i = f i+ – f i–, where fi– and fi+ are the channel band edges, can be evaluated as follows. In the absence of interference from other channels/users (and ignoring electronic noise sources in the devices), the variance of the noise picked up in a band- width of ∆fi may be calculated using the formula σn2 = kBT∆fi, where kB = 1.38 ⋅ 10–23 J/K denotes the Boltzmann constant and T is the system tem- perature in Kelvin. When the leakage from other users is included, the variance of the interfer- ence picked up in the same band will be obtained as (1) where Hi(f) is the frequency response of the receiver filter of the desired channel, Pj is the power of the jth interferer at the transmitter, Gj(f) is the channel response between the jth interferer and the desired channel, and j includes σ inter 2 2 = − +∫∑ ff j j i ji i P H f G f df( ) ( ) The concept of the noise floor is a means for evaluating the effective background noise in heavily utilized parts of the spectrum. Along these lines, interference from SU sources may raise the noise floor for PUs. FARHANG LAYOUT 3/24/08 5:13 PM Page 81 IEEE Communications Magazine • April 200882 all the user signals that have significant contribu- tion. Clearly, the noise and interference variances add up. This leads to the total variance σ2total = σ2n + σ2inter, and the equivalent temperature Ttotal = σ2total/kB∆fi. REDUCING THE INTERFERENCE TEMPERATURE TTOTAL In σ2total = σ2n + σ2inter, σ2n depends only on the temperature of the environment and thus cannot be reduced by means of system design. From Eq. 1, one obvious way to reduce σ2inter is to reduce the output power of the SUs. However, this may not be practical. For instance, consider the situa- tion where an SU that is close to a PU connects to a remote SU base station. In such cases the output power of the SU must remain high simply to maintain connectivity. Moreover, to optimally realize the spectrum opportunities in CR scenar- ios with high differential power between PUs and SUs, a spectral sensor with a very wide dynam- ic range is a necessity. On the other hand, σ2inter is also determined by Hi(f) and Gj(f), and thus may be reduced by a judicious selection of these responses. In par- ticular, Eq. 1 suggests that to minimize the interference temperature, |Hi(f)Gj(f)| should be small for any j ≠ i. In the following sections we show that in a filterbank multicarrier sys- tem, |Hi(f)Gj(f)| can be made arbitrarily small with relatively low computational cost and at virtually no cost of additional bandwidth. In contrast to this, in an OFDM system the reduc- tion of |Hi(f)Gj(f)| can be very expensive from both complexity and bandwidth efficiency per- spectives. LIMITATIONS OF OFDM IN COGNITIVE RADIO SYSTEMS In the following we elaborate on some shortcom- ings of OFDM in CR networks. In OFDM the ith subcarrier signal is characterized by a power spectral density (PSD) of the form Φi(f) = Ksinc2((f – fi)TS) (2) where K is a constant determined by the signal level, sinc is the sinc function defined as sinc(x) = sin(pix)/pix, fi is the center frequency of the subcarrier, and TS is the OFDM symbol dura- tion, which consists of the duration of one FFT block, T, and the guard interval, TG. Assuming that the symbols in different subcarriers are independent of each other, the PSD of an OFDM signal is obtained as (3) where the index i includes all active subcarriers. Because of the relatively large side-lobes of the sinc pulse in Eq. 2, the out-of-band energy generated by an OFDM signal can be significant. In a CR setting, this may result in unacceptable interference to PUs. The sinc shape of the sub- carrier spectra is a consequence of the abrupt transition among successive OFDM symbols. One can avoid the sinc pulse by introducing soft transitions among successive symbols through cyclic extension of each OFDM symbol from TS to (1 + 2β)TS and windowing by a raised cosine shape of the form shown in Fig. 1. The succes- sive OFDM symbols are then overlapped as also shown in Fig. 1. As a result, the effective dura- tion of each OFDM symbol is increased from TS to (1 + β)TS, which constitutes a bandwidth loss of β/(1 + β). In [1] the various choices of β have been examined, and it is concluded that to obtain a reasonable suppression of the out-of-band energy, values of β as large as β = 1 may be needed. To demonstrate this here, in Fig. 2 we present an example of the PSD for various choic- es of β. This clearly shows the large side-lobes of the rectangular window (β = 0) and how the side-lobes decrease in magnitude as β increases. One important observation from Fig. 2 is that even though the raised cosine window is very effective in reducing the side-lobes of the sub- carrier spectra, the side-lobes that are close to the main lobe are still large. This point was also noted in [1], and as a further measure to resolve the problem, a subcarrier deactivation mecha- nism that avoids subcarrier bands near the active PU bands was proposed. This method clearly further reduces the bandwidth efficiency of the CR system. In [1] it is argued that one may trade the use of smaller β (less bandwidth loss due to a raised cosine window) for a larger number of deactivated subcarriers. Another interesting development is by Bran- des et al. [11], where the authors propose a method for further reduction of the side-lobes by assigning nonzero values to the deactivated subcarriers to reduce the magnitude of the side- lobes generated by the data subcarriers. The results presented in [11] report the use of β = 0.2 and can achieve side-lobes at around –60 dB. However, the selection of cancellation subcarri- ers involves a constraint optimization that must be repeated for each OFDM symbol. Another important point that deserves some attention is the interference received by an SU from PUs and other SUs. Weiss et al. [1] acknowledged the issue without proposing any method of reducing it. We recall from the DSL literature that this problem can be solved by applying a window to the received signal prior to passing it to the FFT block for demodulation. The concept of windowing here is somewhat similar to that used at the transmitter. However, differences exist. Figure 3 depicts the method of receiver windowing. When the window is rectan- gular, one picks N samples of a received OFDM symbol after removing the cyclic prefix samples and passes them to an N-point FFT for demodu- lation. To apply a raised cosine window, (1 + Φ Φ( ) ( )f fi i = ∑ n Figure 1. Windowing at the transmitter. βTs βTsTs FARHANG LAYOUT 3/24/08 5:14 PM Page 82 IEEE Communications Magazine • April 2008 83 α)N samples are chosen and windowed, as also shown in Fig. 3. In order to obtain samples of the windowed signal in the frequency domain at the middle of the N subcarrier bands, a Fourier transform has to be applied to (1 + α)N time- domain samples, followed by decimation to N output samples. This can be realized by aliasing in the time domain and applying an N-point FFT. The arrows in Fig. 3 demonstrate how this is performed; essentially, the samples in the shaded areas are added to the windowed sam- ples at the two corners of the time period T. Clearly, application of windowing at the receiver also requires the addition of cyclic prefix and suffix samples, which further reduces the band- width efficiency of OFDM. FILTERBANK MULTICARRIER COMMUNICATION TECHNIQUES Pioneering work on filterbank multicarrier com- munication techniques was done by Chang [5] and Saltzberg [4] in the mid-1960s. Saltzberg showed that by proper design of a transmit pulse shape in a multichannel QAM system, and intro- ducing a half symbol space delay between the in- phase and quadrature components of QAM symbols, it is possible to achieve a baud rate spac- ing between adjacent subcarrier channels and still recover the information symbol, free of intersym- bol (ISI) and intercarrier interference (ICI). This leads to maximum spectral efficiency. Further progress was made by Hirosaki [2], who showed that the transmitter and receiver part of this mod- ulation method could be implemented efficiently in a polyphase/DFT structure. The method was called orthogonally multiplexed QAM (OQAM) in [2]. OQAM has later been referred to as OFDM-OQAM, with the acronym OQAM stand- ing for offset QAM, reflecting the fact that the in- phase and quadrature of each QAM symbol are time offset with respect to each other. In the 1990s advancements in DSL technolo- gy motivated more activity in the development of other filterbank-based multicarrier communica- tion systems to better suit DSL channels. Early development in this area is an American Nation- al Standards Institute (ANSI) contribution by Tzannes et al., which was later expanded and called discrete wavelet multitone (DWMT) [7]. In [8] it was shown that DWMT uses cosine- modulated filterbanks that are more frequently used for signal compression. The name cosine modulated multitone (CMT) was later adopted for this class of modulators. Other interesting advancements that were reported in [8] are: • In CMT, each subcarrier channel transmits a PAM symbol using vestigial sideband (VSB) modulation. • The very particular structure of the subcar- rier signals in CMT leads to a simple blind detection algorithm. Filtered multitone (FMT) is another multicarrier modulation technique specifically developed for DSL applications [6]. As opposed to OFDM- OQAM and CMT, which allow for overlapping of adjacent subcarrier bands, in FMT subcarrier bands are disjoint. It is thus less bandwidth effi- cient than CMT and OFDM-OQAM. SPECTRUM SENSING METHODS In non-parametric spectral estimation, an esti- mate of the PSD of a random process x(n) is obtained by passing x(n) through a bank of nar- rowband bandpass filters and measuring the average output power of these filters. The sim- plest member of this class of spectral estimators is the periodogram spectral estimator (PSE), where a bank of filters whose coefficients are those of the discrete Fourier transform (DFT) are used. The multitaper method (MTM) [10] is one of the most advanced non-parametric spectral esti- mation methods. In MTM, each point of the desired PSD is obtained by averaging the signal power at the output of a set of narrowband (known a