Effects of Bit Allocation on Non-contiguous Multicarrier-based Cognitive Radio Transceivers

Transcription

1 Eects o Bit Allocation on on-contiguous Multicarrier-based Cognitive Radio Transceivers Aleander M. Wglinski Inormation and Telecommunication Technolog Center The Universit o Kansas Lawrence, Kansas alew@ittc.ku.edu Abstract In this paper, we evaluate a cognitive radio transceiver emploing both non-contiguous multicarrier modulation (C-MCM) and adaptive bit allocation. Although C- MCM and bit allocation have potential beneits with respect to enabling dnamic spectrum access (DSA) and increasing throughput, the also require the transmission o overhead inormation between the transmitter and the receiver. To reduce this overhead inormation, operating parameters can be assigned to a block o subcarriers, at the cost o some throughput. The trade-os between subcarrier block size and two dierent bit allocation approaches or several DSA scenarios are assessed in this work. The results show that as percentage o available spectrum decreases, the throughput loss o sstems emploing larger subcarrier block sizes rapidl increases. evertheless, larger block sizes also ield greater reductions in transmission overhead. I. ITRODUCTIO With the demand or additional bandwidth increasing due to eisting and new services, both spectrum polic makers and communication technologists are seeking solutions or this apparent spectrum scarcit. Meanwhile, measurement studies have shown that much o the licensed spectrum is relativel unused across time and requenc [1]. evertheless, current regulator requirements prohibit unlicensed transmissions in these bands, constraining them instead to several heavil populated, intererence-prone requenc bands. To provide the necessar bandwidth required b current and uture wireless services and applications, the Federal Communications Commission (FCC) has commenced work on the concept o unlicensed users borrowing spectrum rom spectrum licensees [2, 3]. This approach to spectral usage is known as dnamic spectrum access (DSA). Simultaneousl, with the rapid evolution o microelectronics, wireless transceivers are becoming more versatile, powerul, and portable. This has enabled the development o sotware-deined radio (SDR) technolog, where the radio transceivers perorm the baseband processing entirel in sotware, e.g., modulation/demodulation. The ease and speed o programming baseband operations in an SDR makes this technolog a prime candidate or DSA networks. SDR transceivers that can rapidl reconigure operating parameters due to changing requirements and conditions are known as cognitive radios [4]. With recent developments in cognitive radio technolog, it is now possible or these sstems to simultaneousl respect the rights o incumbent license holders while providing additional leibilit and access to spectrum. To eploit the advantages o cognitive radio transceivers and enable unlicensed users to transmit in the presence o incumbents license holders, several researchers have proposed a leible modulation technique based on multicarrier modulation that turns o subcarriers which would otherwise interere with incumbent transmissions. This technique is known as non-contiguous multicarrier modulation, or C-MCM [5 7] 1. To urther eploit the leibilit oered b cognitive radio transceivers and C-MCM, bit allocation 2 and other transmission parameter adaptations can be emploed to enhance sstem perormance, such as throughput [9]. However, the design trade-os involved with emploing these techniques have never reall been assessed with respect to computational and implementation compleit, as well as the overhead inormation required or the transmission parameter adaptations. Given the restrictions on hardware resources and computational power o portable, sel-contained cognitive radio implementations, such trade-os need to be determined beore implementation. In this paper, we eamine the design trade-os associated with a cognitive radio transceiver emploing C-MCM transmission and bit allocation in a single user scenario. Speciicall, we will ocus on implementations that ull eploit the leibilit oered b C-MCM using non-uniorm bit allocation and compare them with implementations that attempt to reduce the computational compleit and transmission overhead 3. This paper is organized as ollows: Section II presents an overview o a multicarrier-based cognitive radio transceiver and C-MCM. Section III describes the process o bit allocation. Simulation results and comparisons are presented in Section IV, and several concluding remarks are 1 In the literature, most researchers emplo an eicient orm o multicarrier modulation called orthogonal requenc division multipleing (OFDM) as the basis or C-MCM transmission. For these implementations, C-MCM is usuall reerred to as either non-contiguous OFDM (C-OFDM) or discontiguous OFDM (D-OFDM). 2 The subcarrier signal constellations are adjusted to the prevailing channel conditions in order to achieve some perormance goal, such as throughput optimization [8]. 3 Several techniques to achieve this include using uniorm bit allocation instead o non-uniorm bit allocation, and assigning a signal constellation to a block o subcarriers rather than per subcarrier.

2 Allocation Algorithm Adaptive Bit Demultipleer channel state inormation eedback requenc availabilit eedback subcarrier bit allocation g n ( ) g (1) rn ( ) ( n) hn ( ) ( g 1) ( n) + vn ( ) Secondar User Spectrum Primar User Spectrum (a) Individual subcarriers (i.e., B =1). requenc (a) Transmitter with bit allocation algorithm and the channel. n n+1 n+2 n+3 n+4 n+5 n+6 Channel Estimator & Sounder r( n) ( 1) ( n) ( ) ^ ( n) ^ ( ) ^ ( n) Freq. Domain Equalizer Adaptive Bit Multipleer ^ (1) ( n ) ^ ( n) (b) Receiver with channel estimator and sounder. ulled ulled ulled ulled ulled (b) Subcarrier blocks o size B =4. requenc Fig. 2. Schematic o unlicensed (secondar) users operating in the requenc domain when emploing non-contiguous multicarrier modulation in the presence o transmissions rom incumbent (primar) users. Fig. 1. Schematic o a single user multicarrier-based cognitive radio transceiver operating in the downlink direction, emploing bit allocation, and using eedback rom the channel estimator and sounder. made in Section V. II. SYSTEM FRAMEWORK The general setup or a multicarrier-based cognitive radio transceiver is shown in Fig. 1. The high speed input smbol stream, (n), is demultipleed into streams, with stream i having b i bits per smbol epoch. The value o b i is determined b the allocation algorithm, which uses the subcarrier SR values γ i, i =,..., 1, to compute the subcarrier BER [8]. The subcarrier SR values are computed rom the channel state inormation (CSI) provided b the dataaided channel estimator at the receiver. We onl consider the downlink in this paper, with bit allocation decisions perormed solel at the transmitter. Furthermore, inormation regarding the spectral availabilit across the transmission bandwidth, obtained through channel sounding and spectrum analsis [1, 11], is also used b the transceiver to deactivate subcarriers, i.e., b i =, that can potentiall interere with incumbent transmissions. Once the bit streams are modulated onto one o several signal constellations consisting o M i =2 bi points, the outputs, (i) (n), i =,..., 1, are upsampled b a actor to produce (i) (n), i =,..., 1, and iltered b snthesis ilters g (i) (n), i =,..., 1, beore being summed together, ielding the composite transmit signal, s(n). This signal is transmitted across the channel, where the multipath propagation and additive noise are modelled with channel impulse response h(n) and noise v(n). The received signal, r(n), is separated into the subchannels using the analsis ilters (i) (n), i =,..., 1, downsampled b a actor, equalized using requenc-domain equalizers, demodulated, and then multipleed together to orm the estimate o (n), ˆ(n). Moreover, the receiver uses the subcarrier inormation ŷ (i) (n), i =,..., 1, to generate a channel estimate, as well as to locate and identi non-negligible transmissions within the transmission bandwidth as either spurious intererence/noise or an incumbent user. The identiication process is perormed using one o several spectral analsis techniques (reer to [1, 11] and reerences therein). Once the locations o incumbent transmissions have been obtained, the transceiver then conigures itsel or C-MCM, as described in the net subsection. A. on-contiguous Multicarrier Modulation Given the locations in requenc o spectrum occupied b incumbent users, the goal o the cognitive radio transceiver emploing C-MCM is to deactivate subcarriers that could potentiall interere with these users and transmit over the remaining active subcarriers. Reerring to Fig. 2(a), we observe the spectral usage o the incumbent (primar) and unlicensed (secondar) users. otice how the subcarriers o the unlicensed user are evenl spaced through requenc. Moreover, observe how the subcarriers located in the same vicinit as the incumbent spectrum are deactivated, i.e., nulled, resulting in the non-contiguous characteristic o the multicarrier signal. Although ver leible, the amount o overhead inormation required to indicate whether a subcarrier should be activated or not is large, especiall i this inormation is requentl updated 4. One solution is to activate or deactivate blocks o 4 Given that the spectral occupanc and location o an incumbent transmission are both random, the unlicensed transceiver must be monitoring the spectrum requentl to avoid intererence.

3 subcarriers 5, as shown in Fig. 2(b). In this case, we choose a subcarrier block size o B =4subcarriers across the entire transmission bandwidth o operation. As a result, the amount o overhead inormation is reduced b a actor o B. evertheless, with the reduction in overhead comes the trade-o that the transceiver loses leibilit, resulting in a decrease in throughput. This is due to the act that instead o nulling a single subcarrier, one must deactivate the entire block containing that interering subcarrier. Moreover, the larger the block size B, the greater the chance o having a subcarrier interering with an incumbent user, resulting in having all the subcarriers in that block nulled. On the other hand, the amount o overhead is substantiall reduced. Once the C-MCM transceiver has decided on which subcarriers to activate, bit allocation can be perormed, as will be discussed in the net section. III. ADAPTIVE BIT ALLOCATIO One o the primar advantages o multicarrier modulation is its abilit to transorm a requenc-selective ading channel into a collection o approimatel-lat subchannels. As a result, distortion compensation o the transceiver becomes simpler to perorm. Furthermore, the agilit o the transceiver to tailor its operating parameters to the channel conditions is enhanced due to the resolution o the subcarriers. The subcarrier signal constellation is one operating parameter that can be tailored to the channel conditions. To illustrate, suppose we have a requenc-selective ading channel, a constant transmit power level across all subcarriers, and additive white Gaussian noise. The resulting subcarrier signal-to-noise ratios (SR) and bit error rates (BER) will probabl not be equal or all subcarriers. Moreover, those subcarriers with low SR values/high BER values will dominate the average BER o the overall transceiver. B changing the subcarrier signal constellations, the subcarrier BER values can be changed in order to ield a better average BER or the sstem. The process o changing the subcarrier signal constellations is known as bit allocation. Mathematicall, the process o perorming bit allocation in order to increase the overall throughput o the sstem while ensuring the mean BER, P, is below a speciied mean BER limit, P T, can be deined b the ollowing optimization problem: subject to : b i (1) 1 ma b i i= P = ( 1 i= )/( 1 b i P i i= b i ) P T, (2) where b i is the number o bits per smbol or subcarrier i, P i is the BER or subcarrier i, which is computed rom 5 Although multiband OFDM (MB-OFDM) also groups subcarriers together [5], this is done to reduce the hardware cost o the implementation, instead o the overhead inormation. As a result, this allows or ultra-wideband (UWB) bandwidths in ecess o 5 MHz to be supported b the transceiver. the subcarrier signal-to-noise ratio (SR), γ i, via closed orm epressions [12] 6. As discussed in Section II-A, one o the disadvantages o eploiting the leibilit o multicarrier modulation is the amount o overhead inormation generated. One solution is to perorm uniorm bit allocation. As oppose to non-uniorm bit allocation, where the subcarrier signal constellations can var [8], uniorm bit allocation imposes the additional constraint o b = b 1 =...= b 1 (3) when tring to solve or the objective unction o Eq. (1). Another solution that emplos some o the leibilit oered b multicarrier modulation is to assign a signal constellation to a block o B subcarriers 7. The bit allocation process would assess the average SR o each block o subcarriers, and then select an appropriate signal constellation or each block, insuring that the BER constraint o Eq. (2) is satisied while attempting to increase the sstem throughput in Eq. (1). In the net section, the design trade-os discussed in this paper, e.g., subcarrier block size, uniorm versus non-uniorm bit allocation, are evaluated or an C-MCM cognitive radio transceiver. IV. SIMULATIO RESULTS A. Simulation Parameters In this work, several o the operating parameters rom the IEEE Std a [13] have been emploed in these computer simulations 8. For instance, although the cognitive radio transceiver consists o = 512 subcarriers within a bandwidth o 128 MHz, this is approimatel equivalent to transmitting eight IEEE Std a transmissions in adjacent requenc bands. The BER limit o the transceiver was set to be P T =1 5. Moreover, or the purpose o straightorward comparison, no channel coding was emploed b the sstem. With respect to subcarrier block size, values o B = 1, 8, 16, 32 were used b the transceiver. Furthermore, the unlicensed transceiver was evaluated or dierent percentages o available spectrum. Regarding the dierent bit allocation algorithms studied in this work, two tpes were considered. The non-uniorm bit allocation algorithm proposed b Wglinski, Labeau, and Kabal [8] was emploed, where the allocation was applied to the blocks o B subcarriers. The other tpe was a simple uniorm bit allocation approach applied to all subcarriers, such that largest signal constellation obeing Eq. (2) was chosen. The statistical indoor propagation modeling technique o Saleh and Valenzuela [14], which emplos a Raleigh ading statistic, was used in this work. We used a mean cluster arrival time o 1 µs,ameanraarrivaltimeo1 µs, a 6 In a practical implementation, the BER values would be stored in a look-up table. 7 This approach is similar to the block technique emploing in Section II-A or the reduction o transmission overhead. 8 The operating requenc o the transceiver is 5 GHz, and each subcarrier can emplo M =5signal constellations: BPSK, QPSK, square 16-QAM, square 64-QAM, and null (turned o).

4 Throughput (bits/smbol) Throughput (bits/smbol) Signal to oise Ratio (db) Signal to oise Ratio (db) Fig. 3. Throughput results o a cognitive radio transceiver emploing C- MCM when 1% o the spectrum is available or unlicensed transmission, given block sizes o B subcarriers. ote that uniorm bit allocation (no stars) and non-uniorm bit allocation (with stars) were emploed. Fig. 4. Throughput results o a cognitive radio transceiver emploing C- MCM when 9% o the spectrum is available or unlicensed transmission, given block sizes o B subcarriers. ote that uniorm bit allocation (no stars) and non-uniorm bit allocation (with stars) were emploed. cluster power-deca time constant o 2 µs, and a ra powerdeca time constant o 6 µs. For each time-invariant channel realization, the phsical separation between the transmitter and receiver was varied between 1 m and 6 m 9. The sstem was evaluated at 7 dierent average SR values, and the trials were repeated or dierent channel realizations 1, with the results averaged. B. Throughput Results In Fig. 3, throughput results o a cognitive radio transceiver emploing C-MCM when 1% o the spectrum is available or unlicensed transmission is presented or our dierent subcarrier block sizes, when either uniorm or non-uniorm bit allocation is perormed. We observe that when the sstem emplos uniorm bit allocation, the throughput given the our dierent block sizes are all equivalent. This is due to the act that when 1% o the spectrum is available, all the subcarriers are active and emplo the same signal constellation. As a result, block sizes are irrelevant with respect to throughput. evertheless, block size will make an impact with respect to overhead reduction. When non-uniorm bit allocation is perormed, there are signiicant throughput gains. Although all the curves reach the maimum o 372 bits per smbol 11, the curves or non-uniorm bit allocation achieve greater throughput sooner due to the leibilit o the allocation. Moreover, as the block size decreases in size, increasing the leibilit, the throughput increases rapidl. When 1% o the spectrum is occupied b incumbent transmissions, as shown in Fig. 4, there are some noticeable dierences between the throughput curves here and those in 9 The change in transmitter/receiver separation distance corresponds to an SR change ranging rom 59 db to -11 db. 1 For each channel realization, the locations o incumbent users in the requenc domain is dierent, although the percentage o the occupied bandwidth is the same. 11 This is equal to 512 subcarriers multiplied b 6 bits per subcarrier or 64-QAM modulation. Fig. 3. First, the maimum attainable throughout o 2766 bits per smbol is onl achieved b transceivers using the smallest block size, i.e., B =1subcarrier. Moreover, the more leible non-uniorm bit allocation implementation reaches that maimum throughput signiicantl aster relative to the uniorm bit allocation implementation. Second, as the block sizes get larger, maimum attainable throughput declines substantiall. This is due to the act that i one subcarrier is interering with an incumbent user, the whole block o subcarriers is nulled, resulting in a signiicant throughput penalt. Hence, the larger the block, the greater the penalt. Third, the block-wise nonuniorm bit allocation outperorms the transceiver emploing the uniorm bit allocation, especiall in the mid-range SR values. C. Overhead Reduction and Trade-os Regarding the amount o overhead required or each bit allocation technique, the non-uniorm bit allocation takes 3 bits to represent 5 possible signal constellations. Since each block has the same signal constellation, this translates into 3 /B bits to represent a bit allocation and a subcarrier activit level. On the other hand, or uniorm bit allocation, onl 1 bit is required to indicate the subcarrier activit level per block, and 3 bits or the entire transmission to represent the 5 possible signal constellations. This translates into 1 /B+3 bits to represent an allocation and subcarrier activit. ote that to obtain a straightorward comparison, we assume that no source coding is perormed on the overhead inormation. Thus, the amount o overhead inormation required b the C-MCM cognitive radio transceiver, emploing either uniorm or non-uniorm bit allocation with block sizes o B subcarriers, is given in Fig. 5. It can be observed that a sstem emploing non-uniorm bit allocation generates as much as three times the amount o overhead relative to a sstem emploing uniorm bit allocation. When the transceiver emplos non-uniorm (uniorm) bit allocation, an overhead reduction o 87.5% (87.%), 93.8% (93.2%), and 96.9%

5 Transmission Overhead (bits/coniguration) Subcarrier Size (B) Fig. 5. Amount o overhead transmission per coniguration or the C- MCM cognitive radio transceiver emploing either uniorm or non-uniorm bit allocation with block sizes o B subcarriers. Throughput (bits/smbol) Spectral Availabilit (%) Fig. 6. Maimum-attainable throughput results o a cognitive radio transceiver emploing C-MCM or dierent percentages o spectrum availabilit, given block sizes o B =1, 8, 16, and 32 subcarriers. ote that results were obtained or an SR o 59 db. (96.3%) is achieved when using block sizes o B =8, 16, and 32 subcarriers, relative to an implementation using B = 1. With respect to the maimum-attainable throughput o the cognitive radio transceiver, we observe in Fig. 6 that smaller subcarrier block sizes achieve higher throughput values relative to sstems emploing larger block sizes. ote that both bit allocation algorithms will converge to the same maimumattainable throughput, as shown in Figs. 3 and 4 or high SR values, i.e., an SR o 59 db. However, as the percentage o available spectrum decreases, there is an overall decrease in the maimum-attainable throughput. The decrease is greater or larger subcarrier block sizes. For instance, the throughput loss o a transceiver emploing either bit allocation technique, when the spectral availabilit is 95% (85%), is 29.8% (67.9%), 53.4% (91.3%), and 78.9% (99.4%) or block sizes o B =8, 16, and 32 subcarriers, relative to a sstem with B =1. Thus, when we compare the percentage decrease in overhead to the percentage throughput loss due to decreased leibilit, we can conclude that or spectrum sparsel occupied b incumbent users, the percent gain in reduced overhead is signiicantl greater than the throughput loss due to subcarrier block size. However, as the spectrum ills up with incumbent transmissions, the advantages o establishing blocks o subcarriers quickl diminishes, especiall or large block sizes. V. COCLUSIO We have eamined the throughput perormance o a cognitive radio transceiver emploing C-MCM that uses either uniorm or non-uniorm bit allocation. To reduce overhead inormation and bit allocation algorithm compleit, the transceiver was implemented to assign the same signal constellation and activit level to blocks o subcarriers. The results show that or low spectral occupanc b the incumbent users, the cost o using blocks o subcarriers to reduce overhead was worth it relative to the throughput penalt incurred b using blocks. However, as the incumbent spectral occupanc increases, the beneits in reduced overhead relative to the throughput penalt diminished ver quickl. Thereore, it is recommended that one adaptable parameter to be included in the cognitive radio transceiver emploing C-MCM is an algorithm that decides on a value or the subcarrier block size, which is a unction o the incumbent spectral occupanc. REFERECES [1] Federal Communications Commission, Spectrum polic task orce report, ET Docket o , 22. [2] M. J. Marcus, Unlicensed cognitive sharing o TV spectrum: The controvers at the ederal communications commission, IEEE Commun. Mag., vol. 43, no. 5, pp , Ma 25. [3] Federal Communications Commission, Unlicensed operation in the TV broadcast bands, ET Docket o , 24. [4] J. Mitola, III, Cognitive radio or leible mobile multimedia communications, in Proc. IEEE Int. Wksp. Mobile Multimedia Commun., San Diego, CA, ov. 1999, vol. 1, pp [5] M. P. Wlie-Green, Dnamic spectrum sensing b multiband OFDM radio or intererence mitigation, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [6] J. D. Poston and W. D. Horne, Discontiguous OFDM considerations or dnamic spectrum access in idle TV channels, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [7] H. Tang, Some phsical laer issues o wide-band cognitive radio sstems, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [8] A. M. Wglinski, F. Labeau, and P. Kabal, Bit loading with berconstraint or multicarrier sstems, IEEE Trans. Wireless Commun., vol. 4, no. 4, pp , Jul 25. [9] M. B. Pursle, T.C. Roster, IV, and J. S. Skinner, Protocols or the selection, adjustment, and adaptation o transmission parameters in dnamic spectrum access networks, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [1] F. Weidling, D. Datla, V. Pett, P. Krishnan, and G. J. Minden, A ramework or RF spectrum measurements and analsis, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [11] S. Shankar, C. Cordeiro, and K. Challapali, Spectrum agile radios: Utilization and sensing architectures, in Proc. IEEE Int. Smp. ew Frontiers Dnamic Spectr. Access etworks, Baltimore, MD, ov. 25, vol. 1, pp [12] J. G. Proakis, Digital Communications, McGraw-Hill, ew York, Y, USA, 3rd edition, [13] Institute o Electronics and Electrical Engineers, Wireless LA medium access control (MAC) and phsical laer (PHY) speciications: High-speed phsical laer in the 5 GHz band, IEEE Standard 82.11a, ov [14] A. A. M. Saleh and R. A. Valenzuela, A statistical model or indoor multipath propagation, IEEE J. Select. Areas Commun., vol. 5, no. 2, pp , Feb