An Efficient Filter Banks Based Multicarrier System in Cognitive Radio Networks

Transcription

1 RADIOEGIEERIG, VOL. 9, O. 4, DECEMBER 479 An Effcent Flter Bans Based Multcarrer Syste n Cogntve Rado etwors Yun CUI,, Zhfeng ZHAO,, Honggang ZHAG, Yor-Zhejang Lab for Cogntve Rado and Green Councatons Departent of Inforaton Scence and Electronc Engneerng, Zhejang Unversty Invted paper sygon@zju.edu.cn, zhaozf@zju.edu.cn, honggangzhang@zju.edu.cn Abstract. In cogntve rado technques, OFDM s usually regarded as the physcal layer canddate. However, the weanesses of the OFDM technque,.e., usng plan FFT for spectral analyss, decreased bandwdth effcency due to C (cyclc prefx), hgh out-of-band esson, have been ponted out and the ntroducton of flter bans based ultcarrer (FBMC) syste has been advocated by a nuber of authors. In ths paper, we propose an effcent FBMC syste for cogntve rado networ. At the transtter, we propose a decaton transfor decoposton ethod to elnate the unnecessary zero operatons. At the recever, we utlze the analyss flter bans to sense the spectru bands. In order to conquer the shortages of the tradtonal flter bans, we propose a ultstage analyss flter bans, whch can reduce the coputatonal coplexty whle prove the detecton precson when used to sense the spectru bands. And wth an adaptve threshold schee n the power estator, the threshold can be ept very close to the nose power, whch can ncrease the detecton probablty especally n the condton of low SR. Keywords Cogntve rado, FBMC, transfor decoposton, ultstage flter bans, adaptve threshold.. Introducton Cogntve rado (CR) s a revolutonary ntellgence technology whch can axze the utlzaton of the spectru band by allowng the second users (SU) access the spectru holes wthout causng nterference to prary users. Durng recent years, CR has attracted sgnfcant attenton of the research county and becae a subject of nuerous publcatons ncludng boos, specal conferences, specal ssues, tutorals, and research artcles [], []. The opportunstc nature of CR systes offers bg proses n ters of spectru usage, but generates a lot of constrants. Bascally, such systes ust act qucly to access the spectru and to cease transsson, they ust be relable and robust and they ust provde qualty of servce. And n a lcensed band, the an ssue s the protecton of the prary user, although the coexstence wth other CR users has to be consdered as well, for the sae of global effcency. Ths eans that the prary tas n any CR networ s to dynacally explore the spectru holes and deterne spectru band that can be used wthout causng nterference to other users. The benefts of ultcarrer odulaton n spectru sensng context have already been ephaszed, such as hgh spectral resoluton, coonalty of sensng and councaton functons, flexble way of buldng decson statstcs fro basc observatons wthn the sensng wndow and so on. aturally, OFDM s usually the frst choce when talng about ultcarrer systes. However, the weanesses of the OFDM technque,.e., usng plan FFT for spectral analyss, decreased bandwdth effcency due to C (cyclc prefx), hgh out-of-band esson, have been ponted out and the ntroducton of flter bans based ultcarrer (FBMC) syste has been advocated by a nuber of authors [3], [4]. Fro a transsson perspectve, the FBMC technque has the potental to ncrease the bt rate, due to the reduced guard bands and the absence of the cyclc prefx needed n OFDM. FBMC gves also the possblty to allocate dfferent subcarrers to dfferent non-synchronzed users n a spectrally effcent anner. But the out-of-band esson of FBMC s uch lower than OFDM. Applcaton of flter bans technques n spectru sensng n CR has been studed n references [5], [6]. In ths paper, we propose a ultstage DFT flter bans (MS-DFTFB) enhanced wth an adaptve threshold schee to sense the spectru band. It has low coputatonal coplexty and hgh detecton precson coparng to the detecton ethod based on tradtonal DFTFB. And wth an adaptve threshold ethod n our proposed MS-DFTFB, the threshold can be ept very close to the nose power, whch can ncrease the detecton probablty, especally under the condton of low SR. In cogntve rado, the secondary user (SU) usng ultple carrers OFDM or FBMC technques always need to deactvate a nuber of subcarrers n order to avod nterference to the prary user (U). Ths eans the OFDM or FBMC based cogntve rado transtter wll have a large nuber of zero nputs at the IFFT/FFT od-

2 48 YU CUI, ZHIFEG ZHAO, HOGGAG ZHAG, A EFFICIET FILTER BAKS BASED MULTICARRIER SYSTEM I CR ule f there are any subcarrers need to be deactvated as presented n Fg.. Fro a systeatc perspectve, the IFFT and FFT are the crtcal odules of the OFDM transcever, whch are also the ost coputatonally ntensve blocs n the whole OFDM syste. So, an neffcent FFT/IFFT can consderably waste coputaton power and energy effcency of the overall cogntve rado syste. The flter bans syste often nvolves two processes: separaton of the frequency coponents and recobnaton of the coponents to recover the orgnal sgnal. The separaton process s nown as analyss flter bans and the recobnaton process s nown as synthess flter bans. Analyss and synthess flter bans often appear n pars and satsfy the perfect reconstructon condton whch guarantees the perfect reconstructon of the sgnal. We can utlze synthess flter bans to transt nforaton n FBMC syste as presented n Fg.. In order to avod causng nterference to prary users n cogntve rado networs, we need to deactvate soe subcarrers as shown n Fg.. The deactvaton can be realzed by loadng zeros on the ntended subcarrers whle others are loaded wth odulated coplex sybols at the transtter whch s an M band synthess flter bans. Therefore, the nput ponts ay contan a large nuber of zero ponts. By usng transfor decoposton (TD) ethod, we can elnate the unnecessary zero coputatons, whch can help to prove the coputatonal effcency. Fg.. The usage of a spectru band for second user. Up to now, n the lterature, several ethods have been developed to elnate or reduce the coputaton ntensty f the nput ponts wthn an IFFT/FFT odule have any zeros or the nuber of output ponts acqured s relatvely sall. These ethods are collectvely nown as FFT prunng [7], [8], [9]. Subsequently, Sorensen and Burrus proposed another ethod, naed as transfor decoposton (TD) []. In general, TD can be seen as a odfed Cooley-Tuey FFT where the DFT s decoposed nto two saller DFTs []. In the vew pont of hardware pleentaton, TD ethod s ore effcent and flexble than conventonal FFT prunng. The FFT prunng or TD has already been appled to the OFDM syste nstead of standard FFT algorths n [] and [3], whch both showed sgnfcant reducton n arthetc coputaton. In ths paper, we propose an effcent decaton TD schee whch can reduce the coputatonal coplexty coparng wth the conventonal TD ethod under the condton of sparse nput ponts. The rest of ths paper s organzed as follows: Secton descrbes the decaton TD ethod and the nuerc analyss. Spectru sensng usng our proposed ultstage flter bans s presented n secton 3., followed by an exaple of two stage flter bans used for wreless crophone detecton n IEEE 8. WRA n 3.. Secton 3.3 presents the nuerc analyss and sulaton results. Secton 4 s the adaptve threshold schee n power estator followed by the conclusons n secton 5.. Iproved Transfor Decoposton Method for DFT wth Sparse Input onts n FBMC Transtter M pont IFFT E ) z E ) z EM z ) Fg.. Basc structure of FBMC transtter.. Tradtonal Transfor Decoposton The transfor decoposton uses a xture of a Cooley-Tuey FFT and a coputatonal structure slar to Goertzel s algorth [4]. It s shown to be both ore effcent and ore flexble than prunng. Below, a atheatcal dervaton of TD for nput wth few nonzero ponts s gven. The DFT s defned as X n xnw,,,, n z z () where W n = exp(-jπn/). Assue there are L nonzero nputs and there exst a, whch s the nearest power-oftwo nteger larger than L, dvde and defne Q = /. The ndex n can be wrtten as n n n n,,, Q, n,,,. (), Slarly, the ndex can be decoposed as Q,,, Q,,,,.(3), Substtutng n and n () wth () and (3), the DFT can be rewrtten as X Q n n Q Q xn n W n n n Q n x n n n n W W n. (4)

3 Q Q Q Q Q Q RADIOEGIEERIG, VOL. 9, O. 4, DECEMBER 48 By tang advantage of equaton (5) Q n n n xn n W x,,,, Q (5) n we can rewrtten equaton (4) as (6) n x n n X W,,,, (6) where X X Q. For a gven, equaton (6) can be recognzed as a length FFT, whch can be coputed effcently usng a FFT algorth. As the can range fro to Q -, there are Q length FFT operatons. For each length FFT, we need to acqure x (n ) by usng (5). Snce there are L nonzero nput ponts, the ultplcatons used by (5) wll be L at a gven when n traverses fro to -. When =, ( n n) then W, equaton (5) exsts only addtons. Therefore, the total nuber of ultplcatons cost by (5) wll be (Q )L when ranges fro to Q -. The total nuber of ultplcatons used by TD s gven as u TD Q log Q L. (7). roposed Decaton Transfor Decoposton Method Therenafter, we propose an effcent transfor decoposton ethod whch can reduce the coputatonal coplexty further ore coparng to the tradtonal TD ethod. We denote ths ethod as decaton transfor decoposton (DTD), because the proposed ethod needs to decate the nput x(n) nto two separate sets. The ethod DTD s soewhat le the way fro a DFT to decaton-n-te FFT (DIT-FFT). We frst dvde the nput ponts x(n) nto two groups, one s the set wth even ndex n and the other s the set wth odd n, whch s descrbed n (8) x r xr, x xr, r,,,. (8) Then equaton () can be rewrtten as / r xr W / W xr r / X W, (9) r r / whch dvdes a pont DFT nto two / pont DFT. r r r( /) Because of the syetry of W, whch s / W/ W, / X() can be acqured accordng to the X () and X (), as equaton () shows: X X W X, X X where W X,,,, / r ( ) ( ) / r X x r W,and / r ( ) ( ) / r () X x r W. Instead of usng TD on X(), we perfor TD on X () and X (). Then we use (9) to get X(). Assue there are L nonzero nputs and L nonzero nputs are n the even set x (r) whle L (whch s also L - L ) n the odd set x (r). The TD on X () and X () s the sae as (4) and (5), respectvely. We assue ( ) s the nearest power-of-two nteger larger than L (L ), and Q = /( ), Q = /( ). The nuber of ultplcatons needed for our proposed DTD ethod s gven by () u DTD Q log log 4 Q L Q log Q Q L Q L. L () The coponent / n () s the nuber of ultplcatons that equaton (9) needs to copose X () and X () nto X(). Fg. 3 presents the bloc dagra for the proposed novel TD ethod. The nput x(n) s dvded nto two sets as entoned above, one s x (r) and another s x (r). Then the conventonal TD s carred out on x (r) and x (r) separately. The output appng process s used to get the X () and X () n natural order by utlzng (3). The fnal X() we need s then produced by butterfly operaton as equaton (9) shows. It s the cobnaton of conventonal TD and partal DIT-FFT whch can acheve the coputatonal effcency coparng wth the TD. even xn ( ) odd x () r x () r x() x() x ( ) x () r x ( r ) x() x(3) x ( ) W / W / W W / W ( Q ) / W / ( Q ) / x x( () ~ ) x x( () ~ ) x () ~ x ( ) x x ( () ~ ) x x ( () ~ ) x x () ~ ( ) Q length FFT Q length FFT X() ~ X( ) X() ~ X( ) X X () ~ ( ) X () ~ X ( ) X () ~ X ( ) X X () ~ ( ) X () X () ( X ) X () X () ( X ) W Fg. 3. Bloc dagra of proposed effcent TD ethod for sparse nput ponts DFT..3 uercal Results and Coputatonal Coplexty Analyss X ( ) X () X () X () X( ) In ths secton, we dscuss the coputatonal coplexty of conventonal TD and out proposed DTD ethod n dfferent condtons. As entoned above, the coputatonal coplexty of conventonal TD (CTD) s gven by equaton (7), whle the proposed DTD s coplexty s gven by (). Fg. 4 presents the nuber of ultplcatons CTR-FFT, conventonal TD and our proposed DTD needs under the hypothess that the nonzero nput ponts are all n one sngle set. If the nonzero rato s bgger than.5, then we assue the / nonzero nput ponts are n one set and the reans nonzero nput ponts n another set. In Fg. 4,

4 48 YU CUI, ZHIFEG ZHAO, HOGGAG ZHAG, A EFFICIET FILTER BAKS BASED MULTICARRIER SYSTEM I CR when the nonzero nput ponts s less than /, whch s 5 exactly, we assue the nonzero ponts are all n one set, the odd set x (r), for exaple. In the case the nonzero rato s bgger than.5, we assue there are / ponts n one set, and the reans are n another set. When the nonzero rato s bgger than.5, fro Fg. 4, we can see that the nuber of coplex ultplcatons of CTD s equal to CTR-FFT. Whle the nonzero rato s about.75 the nuber of ultplcatons of DTD and CTR-FFT are equal. When the nonzero rato s less than.5 n Fg. 4, the nuber of coplex ultplcatons DTD needs s uch less than CTD needs. Ths s because there s one set do not need to be operated when the nonzero rato s lees than.5. that CTD and DTD need at dfferent dstrbuton degrees. Fro Fg. 5, we can see that the gap between the CTD curve and DTD curve s becong ore and ore narrower as the ncreasng of dstrbuton degree β. The nuber of ultplcatons that CTD and DTD need are alost equal when β =. Ths eans that when the nonzero nput ponts are unforly dstrbutng, the coputatonal coplexty of DTD and CTD are alost the sae. 3. Multstage Flter Bans for Spectru Sensng n Cogntve Rado etwors The recever of the FBMC syste s an analyss flter bans as shown n Fg. 6. And we can utlze the analyss flter bans (AFB) to sense the spectru band. In Fg. 6, the analyss flter bans utlze the polyphase structure and IFFT. We nae ths AFB as DFT flter bans (DFTFB). A tradtonal M band DFTFB dvdes the spectru band nto M subbands and by calculatng the energy of each subband we can now whether ths subband s occuped or not. The detecton precson s drectly affected by the paraeter M, whch s the nuber of subbands of the M band flter bans. If we want hgher detecton precson, we need to ncrease the value of M, whch wll cause the rapd ncreasng of coputatonal coplexty. Fg. 4. The nuber of coplex ultplcatons that CTR- FFT, conventonal TD (CTD) and our proposed ethod (DTD) need under the hypothess that entoned above. z E ) z E ) z M pont IFFT d ^[ ] z E M z ) Fg. 6. Basc structure of FBMC recever. 3. roposed Multstage Analyss Flter Bans Fg. 5. The nuber of coplex ultplcatons that DTD needs at dfferent dstrbuton degrees. When the nonzero nput ponts are arbtrary dstrbutng n the two sets, the result wll be dfferent fro Fg. 5, obvously. We use the dstrbuton degree to denote the randoness of the nonzero nput ponts as β = L /L, L s the nuber of nonzero ponts n the even set x (r) whle L s the nuber n set x (r). Obvously, Fg. 4 s the scenaro that β =. Fg. 5 presents the nuber of ultplcatons In order to reduce the coputatonal coplexty whle prove the detecton precson, we dvde the tradtonal DFTFB nto stages. Stage : We frst use an M band DFTFB to sense the spectru that we nterest n. If there are narrow band users exstng n the spectru band and the detecton result s not precse enough, we carry out stage. Stage : We use an M band DFTFB to detect the narrow band user we nterest n (we use x n (n) to denote our target sgnal below) based on the result of stage. We dvde the subband whch has detected x n (n) n stage nto

5 RADIOEGIEERIG, VOL. 9, O. 4, DECEMBER 483 M subbands. These M subbands can buld up an M DFTFB, whch has uch narrower subband coparng wth M DFTFB. If the narrow band sgnal spans two or ore subbands, for exaple SU, as presented n Fg. 7, we dvde these adjacent subbands nto M subbands and then buld up an M band DFTFB. If the detecton precson s stll not good enough, we can carry out stage 3, stage 4,... stage, untl the detecton precson acheves our requreent. The process s just as sae as stage to stage. Before we carry out our analyss on MS-DFTFB, we need to ae soe useful defntons. Assue the bandwdth of the spectru we want to sense s W and the frequency of the spectru band ranges fro to W. (In fact, by usng a down converson, we can change any spectru band nto base band.) We defne S as the nuber of subbands that x n (n) spans n the M band DFTFB of stage, =,,, -. In Fg. 7, for exaple, S = for SU and S = for SU. s the nuber of total stages of MS- DFTFB. Obvously, the bandwdth of subband n M band DFTFB s W /M. The bandwdth of the subband n the th stage s M band DFTFB (W sub) s gven by M M,,3, Wsub S W /,. () The center frequency of x n (n) (f cx) s decded by the subbands that detected x n (n) n the th stage, whch are assued th to the ( + S - ) th subbands of the M band DFTFB, as presented n equaton (3) W S fcx f,,,, M where f s the start frequency of the th subband. (3) s. In order to sense the th to ( + S - - ) th subbands of the M - band DFTFB, we need to odulate the center frequency of M band DFTFB onto the center frequency of those subbands of M - band DFTFB. Ths can be realzed by ultplyng a odulaton coponent after each polyphase coponent E (z), =,,,M - n Fg. 6. In the th stage, the odulaton coponent (C ) s defned as C exp( j fm j( ) W / ), where sub f M s the start frequency of the flter bans n the ( - ) th stage. The whole structure of our proposed schee s presented n Fg. 8(a). There are stages and each stage (except stage ) has two nput flows. One s the detecton nforaton, whch s used to set the odulaton coponent. The other s the SU sgnal (x n (n)) we want to detect fro the antenna. Fg. 8(b) s the structure of the M band DFTFB n the th stage. In Fg. 8(b), C exp( j fm j( ) W / ) s the odulaton coponent of the th sub stage. z xn xn ( n) ( n) E ( ) z M E ( ) z M exp( j f j( ) W / ) sub M M f f f f f M f z x ( n M ) n E M ) z ym ( ) n f ( ), f f, f ( f ) M, f Fg. 8. (a) Bref structure of the proposed ultstage DFT flter bans. (b) Structure of DFT flter bans wth odulaton coponent n stage L. Fg.7. The schee of proposed ultstage DFTFB. The detaled ethod of dvdng the nterested subband(s) of M - ( > ) band DFTFB nto M subbands and buld up an M band DFTFB s as followng ( th stage): We assue the lowest ndex of the subbands that have detected x n (n) s, whch eans the x n (n) spans fro the th to the ( + S - - ) th subbands of the M - band DFTFB. Then we dvde these S - subbands nto M subbands and buld up an M band DFTFB. We assue the types of the prototype flters of M =,,, band DFTFB are sae except the bandwdth. The ntal start frequency of M band DFTFB 3. An Exaple: Wreless Mcrophone Detecton n IEEE 8. WRA Usng roposed Multstage AFB In 8. WRA, at any te when there s wreless crophone (WM) appearng n TV channel, the whole channel of 6 MHz bandwdth should be evacuated edately for nterference avodance. In order to avod nterference to WM when other users usng the adjacent TV channel, we need to now the precse locaton of WM. Usually, the bandwdth of WM s Hz or less. Tradtonal DFTFB dvde the whole 6 MHz TV channel nto 3 or ore subbands. By detectng the energy of each subband,

6 484 YU CUI, ZHIFEG ZHAO, HOGGAG ZHAG, A EFFICIET FILTER BAKS BASED MULTICARRIER SYSTEM I CR we can locate the occuped frequency of WM. In ths secton, consderng the coplexty of hardware pleentaton, we choose a two stage DFTDB (TS-DFTFB) as entoned n 3. to sense the 6 MHz TV channel. The whole archtecture of two stage flter bans s presented n Fg. 9. z z z z x( n) xn ( ) x( n) xn ( ) xn ( ) xn ( ) E E z ( ) E z ( ) E z ( ) E z E z ( ) ( ) z ( ) e j pont IDFT pont IDFT y ( n) y ( n) Fg. 9. Archtecture of the two stage flter bans. ower Estator ower Estator We use a RF odule followed by an ADC to saple the recevng WM sgnal. A rough detecton s carred out by the band DFTFB at the stage, whch estates the output power of each subband. Then the odulaton coponent s set up by the SET odule accordng to the rough detecton result. A precse result wll be obtaned durng the second stage that reles on uch narrower subband to detect the WM wth ncreased resoluton. The power estator odule s used to calculate the power of each subband and adaptvely settng the threshold. More detal of the power estator odule wll be descrbed n secton 4. We assue the WM users are slowly changng so that there s no need to save the data fro ADC nto a buffer for precse detecton n stage. In addton, there s no need to further dvde the band DFTFB nto uch narrower DFTFB le stage to stage n ths paper, because we consder that the detecton result s relatvely precse enough for the WM, coparng wth the 6 MHz TV channel, and the hardware pleentaton coplexty wll ncrease f we dvde the DFTFB nto 3 or ore stages. 3.3 uercal Analyss and Sulaton Results As ultplcaton s the ost coplex operaton n flterng, we use the nuber of coplex ultplcaton to estate the coputatonal coplexty. In ths secton, we do not consder the coplexty of the power estator odule, because the adaptve threshold schee s perodc and n a relevant long te, we can use the sae threshold and do not need to change t. Correspondng to the tradtonal DFTFB (t-dftfb) entoned above, the nuber of coplex ultplcaton s gven by M n t DFTFB lh n log M (4) where l(h (n)) s the ponts of the prototype flter, M s the nuber of total subbands. In (4), M s a power-of-two nteger. In coparson, the nuber of coplex ultplcaton of the TS-DFTFB s gven by: n TS DFTFB l l h n log h n log (5) where l(h (n)), =,, s the ponts of prototype flters of the, =, band DFTFB. Slar to equaton (4), and are both power-of-two ntegers. The nuber of ultplcatons of the band DFTFB s addng another l(h (n)) copared wth band DFTFB n (5). Ths s because the odulaton at the stage costs addtonal l(h (n)) coplex ultplcatons. The total M ponts of M El ( z ) s just the length of the prototype flter l n Fg. 9, and the nuber of coplex ultplcatons cost on odulatng band DFTFB to the center frequency of WM that we get n stage s just l(h (n)). In order to avod alasng n the M band DFTFB, t s necessary to ensure the ponts of prototype flter be bgger than M. In ths paper, we choose the length of prototype flter γ tes as the total bands of DFTFB, naely l(h (n)) = γm, γ. When we use tradtonal DFTFB or TS-DFTFB to detect the WM, f the WM spans two or ore adjacent flters, we have to choose the center frequency of those flters as the WM s center frequency. Obvously, the bgger the value of M or s, the saller the detecton error wll be. Fg. (a) presents the sulaton results of detecton error of the 3 band tradtonal DFTFB and the TS-DFTFB wth coeffcents of = 8, = 4. A sequence of wreless crophones wth rando center frequences s used to carry out our sulaton. We assue that whenever there s a WM appearng n a specfc TV channel, t can be detected edately. Fro Fg. (a), we can observe that, the detecton precson of tradtonal DFTFB and TS- DFTFB s alost sae when M =. Fg. (b) presents the nuber of coplex ultplcatons that tradtonal DFTFB and TS-DFTFB needed when the value of M and s equal. We can fnd that, when the detecton precson s sae, the nuber of ultplcatons of TS- DFTFB s saller than tradtonal DFTFB.

7 RADIOEGIEERIG, VOL. 9, O. 4, DECEMBER 485 Fg.. (a) Detecton error of 3 band t-dftfb and TS- DFTFB. (b) uber of ultplcatons of 3 band t- DFTFB and TS DFTFB. 4. ower Estator wth Adaptve Threshold In ths paper, we use an adaptve threshold algorth to deterne whether there s WM n TV channel n secton 3. (t obvously can be used n other stuaton of spectru sensng), whch s the AT odule n Fg.. The an dea s cong fro reference [5], but we ade soe odfcaton n our paper. The archtecture n dashed rectangle n Fg. s the detaled structure of power estator odule. The output of each subband of the band DFTFB γ (n), =,,,- s frstly operated by γ (n) /, whch s n order to calculate the power of each subband. Then the subband power coparson (SC) odule s used to copare the power of each subband wth the threshold set by the adaptve threshold (AT) odule. Whether there s WM n TV channel as well as the average nose power p n can be acqured n ths odule. (If several contnuous subbands have bgger output power than the threshold, the SC odule wll decde these subbands are occuped by WM, and the average power of nose p n s the average power of the reanng subbands.) The WM detecton nforaton s just the output of our proposed power estator and the average nose power p n s sent to the adaptve threshold (AT) odule n order to get the next perod s threshold value. y ( ) n p n Fg.. Detaled power estator odule n Fg. 9. The ethod [5] we choose to adaptvely set the threshold s as followng: p p (6) th where p th () s the threshold n the th perod that set by AT, and p th can be adaptvely set accordng to the spectru envronent and the whte Guasson nose (WG). In order to reduce false alar, p th should be bg enough to ensure the threshold s bgger than the nose. Whle n the other hand, p th should be properly set to ensure the threshold s saller than the sgnal, whch can reduce the probablty of leaage alar. Another coponent β() s presented n (7), where the odfcaton s ade n. Cp pn,,3,, (7) where p n ( - ) s the average nose power n the ( ) th perod (n reference [5], p n () wll be nstead of p n ( - )), α and Cp are two adaptve paraeters. α anly affects the tendency of β() and C p can control the degree that β() affected by the nose. We usually set Cp to f the nose s varyng very slowly. The average nose power p n ( - ) can be acqured accordng to the result of the ( ) th perod s detecton result and can be used to end the threshold n the th perod. Ths s the an otvaton that we use p n (-) nstead of p n () n our paper. th Fg.. (a) Threshold and AWG curves, where the ntal threshold s varyng to the actual nose power. (b) Threshold and AWG curves, where the ntal threshold s uch bgger than the nose power. (The values of adaptve paraeters are α = 6, C p =.) We can use the threshold p th () to deterne whether there s WM n TV channel by coparng the power of each subband wth the threshold n the th perod. The value of threshold exported to SC odule fro AT odule wll antan unchangng n the whole perod. Ths eans that we should set the threshold agan after a perod accordng to the spectru envronent, whch s exactly the average nose power p n n ths paper. When we use the adaptve threshold ethod n (6) and (7), we need an ntal threshold p th (), whch s an eprcal value. Even f the ntal threshold we set s bgger or saller than the actual

8 486 YU CUI, ZHIFEG ZHAO, HOGGAG ZHAG, A EFFICIET FILTER BAKS BASED MULTICARRIER SYSTEM I CR nose power, the followng thresholds wll be adaptvely aended to the actual nose power after several perods, just as presented n Fg.. We can see that the adaptve threshold curve s uch soother than the nose power curve fro Fg. (a). And wth a proper value of p th, we can ensure the threshold alost always bgger than the nose power n order to reduce the probablty of false alar. Fg. (b) s the stuaton that ntal threshold p th () s uch bgger than the actual nose power. We can fnd that after several perods, the threshold s very close to the nose power just as n Fg. (a). 5. Conclusons In ths paper, we propose an effcent FBMC syste for cogntve rado networs. At the transtter, we propose a decaton transfor decoposton ethod to elnate the unnecessary zero operatons. At the recever, we utlze the analyss flter bans to sense the spectru bands. In order to conquer the shortages of the tradtonal flter bans, we propose a ultstage flter bans, whch can reduce the coputatonal coplexty whle prove the detecton precson when used to sense the spectru bands. Ths schee has been analyzed and tested through sulatons on wreless crophone detecton n IEEE 8. WRA. The sulaton results also deonstrate our theoretcal analyss. Besdes, we also use an adaptve threshold schee to deterne whether there s prary user n a spectru band. Sulaton results show that the adaptve ethod can eep the threshold close to the nose power, even f the ntal threshold s uch bgger or saller than the actual nose power, whch can ncrease the detecton probablty especally n low SR. Acnowledgeent Ths paper s supported by the Key roject of the Educaton Departent of Zhejang rovnce (o. 747), the Key roject of the Offce of Scence and Technology of Zhejang rovnce (o. 8C5-3), and the Key roject of the Mnstry of Industry and Inforaton Technology of Chna ( Research on Broadband Wreless Access Systes Archtecture and Fraewor for Supportng Broadcastng Servces o.9zx35-4). References [] MITOLA, J., MAGUIRE, G. Q. Cogntve rados: ang software rados ore personal. IEEE ersonal Councatons, Aug. 999, vol. 6, no. 4, p [] HAYKI, S. Cogntve rado: bran-epowered wreless councatons. IEEE Journal of Selected Areas n Councatons, Feb. 5, vol. 3, no., p.-. [3] ZHAG, Q., KOKKELER, A. B. J, SMIT, G. J.M. An oversapled flter ban ultcarrer syste for Cogntve Rado. In IEEE ersonal, Indoor and Moble Rado Councatons (IMRC). Cannes (France), 8, p. -5. [4] WALDHAUSER, D. S., BALTAR, L. G., OSSEK, J. A. Coparson of flter ban based ultcarrer systes wth OFDM. In IEEE Asa acfc Conference on Crcuts and Systes (ACCAS). Sngapore, 6, p [5] BEHROUZ, F. B. Flter ban spectru sensng for cogntve rados. IEEE Transactons on Sgnal rocessng, May 8, vol. 56, p [6] SHEIKH. F., BIG, B. Cogntve spectru sensng and detecton usng polyphase DFT flter bans. In roceedngs of IEEE Consuer Councatons and etworng Conference (CCC 8). Las Vegas (USA), Jan. 8, p [7] MARKEL, J. FFT prunng. IEEE Transactons on Audo and Electroacoustcs, 97, vol. 9, p [8] SKIER, D. runng the decaton n te FFT algorth. IEEE Transactons on Acoustcs, Speech and Sgnal rocessng, 976, vol. 4, p [9] ALVES, R. G., OSORIO,. L., SWAMY, M.. S. General FFT prunng algorth. In roceedngs of the 43rd IEEE Mdwest Syposu on Crcuts and Systes. Lansng (MI, USA),, vol. 3, p [] SORESE, H. V., BURRUS, C. S. Effcent coputaton of the DFT wth only a subset of nput or output ponts. IEEE Transactons on Sgnal rocessng, 993, vol. 4, p [] COOLEY, J. W., TUKEY, J. W. An algorth for achne coputaton of coplex Fourer seres. Math. Coput., Apr. 965, vol. 9, p [] SHOUSHEG HE, TORKELSO, M. Coputng partal DFT for cob spectru evaluaton. IEEE Sgnal rocessng Letters, June 996. [3] ZHAG, Q., KOKKELER, A. B. J, SMIT, G. J.M. An effcent FFT for OFDM based cogntve rado on a reconfgurable archtecture. In IEEE Internatonal Conference on Councaton. Glasgow (UK), June 7. [4] GOERTZEL, G. An algorth for the evaluaton of fnte trgonoetrc seres. Jan. 958, Aer. Math. Monthly, vol. 65, p [5] MIJIA ZHAO, Study on ult-band, ult-rate, ult-ode software rado recevers. h.d Thess, Zhejang Unversty, 3. About Authors Yun CUI s a Master student at the Departent of Inforaton Scence and Electronc Engneerng, Zhejang Unversty, Chna. Hs research nterests are cogntve rado and ultcarrer systes, anly focusng on flter bans based ultcarrer syste n cogntve rado networs. In the area of spectru sensng based on flter bans technques, he has one patent applcaton, and also has publshed several papers. Zhfeng ZHAO s an Assocate rofessor at the Departent of Inforaton Scence and Electronc Engneerng, Zhejang Unversty, Chna. He receved the h.d. degree n Councaton and Inforaton Syste fro the LA Unversty of Scence and Technology, anjng, Chna, n

9 RADIOEGIEERIG, VOL. 9, O. 4, DECEMBER 487. ror to that, he receved the Master degree of Councaton and Inforaton Syste n 999 and Bachelor degree of Coputer Scence n 996, fro the LA Unversty of Scence and Technology, respectvely. Fro Septeber to Deceber 4, he acted as a postdoctoral researcher at the Zhejang Unversty. Fro January 5 to August 6, he acted as a senor researcher at the LA Unversty of Scence and Technology, anjng, Chna, where he perfored research and developent on advanced energy-effcent wreless router, Ad Hoc networ sulator and cogntve esh networng test-bed. Hs research area ncludes cogntve rado, wreless ulthop networs (Ad Hoc, Mesh, WS, etc.), wreless ulteda networ and Green Councatons. Honggang ZHAG s a Full rofessor at the Departent of Inforaton Scence and Electronc Engneerng, Zhejang Unversty, Chna. He receved the h.d. degree n Electrcal Engneerng fro Kagosha Unversty, Japan, n 999. Fro October 999 to March, he was wth the Shn-Kawasa Research Center, Telecouncatons Advanceent Organzaton (TAO) of Japan, as a TAO Research Fellow. Fro Aprl to oveber, he joned the TOYOTA IT Center. Fro Deceber to August 4, he has been wth the UWB (Ultra-Wdeband) Research Consortu, at the Councatons Research Laboratory (CRL) and atonal Insttute of Inforaton and Councatons Technology (ICT) of Japan. Fro Septeber 4 to February 8, he has been wth CREATE-ET (Trento, Italy) where he led ts wreless teas n explorng cogntve rado and ts ntegraton wth Ultra-Wdeband technologes for open-spectru wreless councatons and networs evoluton. Hs research nterests are n Ultra-Wdeband wreless councatons, cogntve rado, broadband networs ntegraton, and Green Councatons.