A ovel Algoithm fo Blind Adaptive Recognition between 8-PSK and /4-Shifted QPSK Modulated Signal fo Softwae Defined Radio Application A. Abaviani, M. Soleimani and V. Tabatabavakili Electical Engineeing Depatment, Ian Univeity of Science and Technology, Tehan, Ian Email: abaviani@ieee.og Abtact- In Thi pape, we popoe a new method to ecognizing and ditinguihing among 8-PSK modulated ignal and /4- hifted QPSK, fom the ample of eceived noiy and faded PSK ignal. The popoed algoithm i baed on computation of elative Euclidean ditance and elative phae diffeence among the eceived ymbol in a fame. Pio to ecognition, the Contant Modulu (CM) equalization i pefomed. Baed on imulation eult, thi method can pefom ecognition tak with high accuacy even at low Signal to oie Ratio (SR) value and employing faily mall numbe of data ample. To the bet of ou knowledge in compaion with othe method of ditinguihing among QPSK vaiant, thi algoithm do well pecially in pactical ytem which have a limited time to make the deciion and ue mall numbe of ample of eceiving ignal. Keywod- Automatic Blind Modulation Recognition (ABMR); Mean Squae Eo (MSE); Contant Modulu Algoithm (CMA); Tapped Delay Line filte (TDL); oftwae defined adio (SDR); elative Euclidean ditance; elative phae diffeence I. ITRODUCTIO Automatic modulation ecognition (AMR) ha become moe impotant with the iing development in oftwae defined adio (SDR) ytem. In SDR ytem we could apply automatic ecognize befoe the demodulation block eulting a obutly handle of multiple modulation in a ingle SDR ytem. Thu, modulation ecognition i an impotant iue fo uch ytem. QPSK, OQPSK and /4-Shifted QPSK ae the ecommended modulation type fo a lage numbe of mobile and wiele communication tandad (uch a IS-54, IS-95, 802.11 and etc). To ecove the meage tuly, a pactical SDR ytem would equie to automatically ditinguihing among uch vaiant of QPSK [1]. Modulation type claifie play an impotant ole in ome communication application uch a ignal confimation, intefeence identification, Suveillance, electonic wafae and electonic counte-counte meaue. Many technique have been epoted in liteatue fo AMR. In a epot by Weave, Cole, and Kumland [2], ealy wok on analog modulation ecognition can be found. In the aea of ecognition of digitally modulated ignal, the pape by Liedtke [3] i a well-known ealy wok. He peented eult baed on a tatitical analyi of vaiou ignal paamete to diciminate between amplitude hift keying (ASK), Fequency Shift Keying (FSK), and Phae Shift Keying (PSK) ignal. Vaiou technique uch a Atificial eual etwok (A) [4], contellation hape [5], Statitical moment matix method [6], maximum likelihood [7], and thei combination have been ued fo AMR. In addition, a few numbe of thee method ae thehold-baed technique to etimate modulation cheme [3, 4]. Fo uch cheme, the thehold level become SR dependant o thehold etting i difficult unde vaiable SR cenaio. A method baed on Mean Squae Eo (MSE) deciion ule to ecognize diffeent M-PSK ignal have been developed in [8]. In thi method the autho compute MSE between the pototype meage point toed in the eceive libay and the eceived ignal point. Claification i made by computing the diffeence in MSE of diffeent PSK ignal againt pecified thehold value. In [1], ame autho popoed a method fo ditinguihing among diffeent type of QPSK vaiant uch a QPSK, OQPSK and /4-Shifted QPSK ignal among Additive White Gauian oie (AWG) and multi-path fading channel. In thi pape, we popoe a novel algoithm to ditinguih between 8-PSK and /4-Shifted QPSK ignal efficiently in AWG and multi-path fading channel. Accoding to the high pefomance of /4-Shifted QPSK modulation, lage numbe of pactical ytem uch a atellite communication ytem, diffeent kind of wiele ytem like IEEE 802.11, WiMax ytem and GSM mobile communication ytem have been employed thi type of modulation. Thee ae two pinciple fo geat pefomance of /4-Shifted QPSK ignal: 1) Aynchonou detection ability 2) Le amplitude vaiety than QPSK ignal becaue of le phae fluctuation. On the othe hand, becaue of low enegy level and good pefomance, many wiele and othe communication ytem may ue the 8-PSK modulation fo tanmiion the data. A we know the contellation hape of thee two modulation ae quite imila. So thee i an impotant poblem to ditinguih /4-Shifted QPSK ignal fom 8-PSK ignal. 978-1-4244-3424-4/09/$25.00 2009 IEEE
The et of the pape i oganized a follow; In ection II the effect of channel on contellation point of the eceived ignal ha been invetigated. In ection III we biefly decibe the method developed in [1] fo blind ecognition of diffeent M-PSK ignal. The popoed method fo ecognition of 8-PSK and /4-Shifted QPSK vaiant with imulation eult ae peented in detail in ection IV. Concluion ae tated in ection V. II. COSTELLATIO POITS PASSED FROM AWG AD FADIG CHAELS The eceived band-pa ignal in the k-th ignaling inteval in the AWG channel may be witten a tk (, ) m (, t k) ntk (, ) kt t ( k 1) T (1) Whee T i ymbol duation, m (t) i the meage wavefom coeponding to the M-PSK ymbol m, m = 1, 2, 3...M. Auming pefect caie ynchonization and timing ecovey and employing I-Q demodulation we get k ( ) [ I( k), Q( k)] [ mi ni ( k), mq nq ( k)] (2) Thu in the ignal pace the eceived ignal point wande aound ignal point in a completely andom fahion, in the ene it may lie anywhee inide a Gauian ditibuted noie cloud centeed on the meage point. The effect of AWG on ignal point fo MPSK ignal at the eceive i hown in Fig. 1(b). Fo wiele communication cenaio, in addition to AWG, thee will be the effect of multi-path fading. Multipath fading channel can be modeled by a Tapped Delay Line (TDL) [1]: the tet ignal i convolved with the impule epone of the TDL to account fo the effect of fading induced by the channel. The TDL paamete ae choen coeponding to powe delay pofile of phyical channel. We have conideed a fequency elective channel in pefoming imulation. Fig. 1(c) and Fig. 1(d) epectively how the faded eceived ignal contellation and equalized ignal contellation afte CM equalization. Figue 1. Effect of oie and Fading on QPSK and 8-PSK contellation at SR=10 db, (A) Pototype ignal point, (b) Received noiy ignal contellation, (c) Received noiy and faded ignal contellation (d) Equalized contellation. III. METHOD FOR AUTOMATIC BLID MODULATIO RECOGITIO In thi ection, we biefly dicu the Automatic Blind Modulation Recognition (ABMR) algoithm developed in [1, 8], fo etimating modulation cheme baed on MSE citeion. Thi dicuion will be helpful in undetanding the popoed algoithm fo ditinguihing QPSK vaiant. A equence of eceived ignal ample { ( k )}, k = 1, 2..., ae collected at demodulato output. Uing thi equence, we check how cloely the eceived ignal ample "match" with each of the pototype contellation available at the eceive libay. The degee of "cloene" o "match" i meaued in tem of a Mean Squae Eo powe defined a 1 2 q MSE ( M ) Dk, M, M 2, q 1,2,... Whee k, m k 1 Dk, M min{ ( k) m }, m 1, 2,... m min{ d } m The computation of D k. M can be implified by confining the each to that quadant in which lie. Fo example, a hown in Fig. 2, a (k) lie in fit quadant (Q 1 ), we need to compute d, 1, k, 2 only the ditance k d d 3 and k, D k, 8 to find. Recognition cheme make the following obevation: Lowe-ode PSK contellation ae ub-et of the higheode PSK cheme; theefoe, when lowe-ode PSK ymbol ae tanmitted, the eceived ignal equence { ( k)} will find a "match" not only with the coeponding pototype contellation, it will alo "match" with the highe-ode contellation (with moe o le the ame degee of accuacy). If the tanmitted ignal i BPSK; the eceived ignal point will be catteed aound the ymbol 2 and 6 hown in Fig. 2. (a) Majoity of the point will be confined in the fit and the thid quadant (Q 1 and Q 3 ) epecially at high SR. The contibution of thee point towad MSE powe will be the ame in both BPSK and QPSK, i.e. MSE (2) MSE (4), k ( ) Q Q 1 3 Howeve, thi ame et of point will eult in a lightly lowe MSE when matched to 8-PSK a ome of thee point will have cloe match to 8-PSK ymbol 1 o 3 and 5 o 7 hown in Fig 2. Thu, MSE (8) MSE (2), MSE (4), ( k ) Q Q 1 3 (b) Fo a mall faction of the eceived point which lie in Q 2 and Q 4, thei match with the BPSK pototype will be pope (the neaet ymbol being 2 and 6 ) a compaed to (3) (4)
Figue 2. Euclidean ditance vecto calculation fo M-PSK ignal. QPSK pototype (neaet ymbol 4 and 8 ) and 8-PSK (neaet ymbol 3, 4, 5 and 7, 8, 1 ). Thu, MSE (8) MSE (4) MSE (2), k ( ) Q Q 2 4 Concluion: when BPSK i tanmitted, at any SR, we hall find MSE(8) < MSE(4) < MSE(2) At high SR, the diffeence in MSE ae negligibly mall; only at low SR, the diffeence ae ditinguihable If QPSK i tanmitted; ow { ( k )} ae catteed aound the fou ymbol 2, 4, 6, 8. It follow that { ( k)} will match well with QPSK and 8-PSK pototype while thee will be lage mimatch with BPSK pototype. Thu, MSE(2) > MSE(4), MSE(8) at all SR MSE(8) MSE(4) at high SR MSE(8) < MSE(4) at low SR If 8-PSK i tanmitted, following imila eaoning we conclude: MSE(2) > MSE(4) > MSE(8) at all SR. ow we eview the peented appoach in [1] to identify QPSK ignal vaiant, uch a QPSK, OQPSK and /4-Shifted QPSK. OQPSK ignaling i imila to QPSK ignaling, except the time alignment of the odd and even bit team by halfymbol peiod. Thi implie that the maximum phae hift of the tanmitted ignal at any given time intant i limited to ±90 degee. The contellation diagam of QPSK and OQPSK i hown in Fig. 3(a) and Fig. 3(b) epectively. It i obeved that, fo QPSK ignal thee i a bidiectional phae tanition fom ignal point 1 to 3 o 2 to 4 and vice vea. Hence the elative Euclidian ditance denoted by D between the adjacent eceived ignal ymbol in a fame fo QPSK i geate than OQPSK ignal. Theefoe the ignal which ha geate D will be booted and the ignal which ha malle D will be minimized, with highe ode even powe moment. The moment M of eceived ignal ymbol in a fame i epeented a Figue 3. Contellation diagam of a QPSK and OQPSK ignal. 1 M D ( k ) k 1 4 Whee D (k) i elative Euclidean ditance between eceived ymbol in a fame length of ymbol epeented a, D ( i) ( j) 1 k, 1 i 1 and j i 1 In a /4-Shifted QPSK modulato, contellation i fomed by upeimpoing two QPSK ignal contellation offet by /4 elative to each othe, eulting in eight ignal phae. Duing each ymbol peiod a phae angle fom only one of two QPSK contellation i tanmitted. Thi eult in maximum phae tanient of 1350. The above mentioned contellation baed ABMR algoithm doe not claify /4-Shifted QPSK and 8- PSK ignal, becaue both cheme have eight contellation point. An appoach which developed to claify /4-Shifted QPSK and 8-PSK ignal depend on elative phae of eceived ymbol in a fame [1]. Fig. 4(a) and Fig. 4(b) depict the elative phae ditibution of 8-PSK and /4-Shifted QPSK ignal epectively. The elative phae diffeence denoted by P between each eceived ymbol in a fame i epeented a V ( i) tan 1 Q P ( ) i VI i V ( k) ( i) ( j),, 1 i 1 1 k, j i 1 Figue 4. Relative phae ditibution of a 8-PSK and /4-Shifted QPSK ignal. (5) (6)
The elative phae ditibution of 8-PSK i 2 3 4 5 6 7,,,,, 0, 8 8 8 8 8 8 and 8. The elative phae ditibution of /4-3 5 7,, Shifted QPSK i 0, 8 8 8 and 8.It i obeved that, without any noie and fading, fo /4-Shifted QPSK ignal thee i no elative phae ditibution of 2 4, 6 8 8 and 8. Hence, to identify the modulation type autho have placed thee counte C1, C2 and C3 at 2 4, 6 8 8 and 8 phae location epectively. The baic function of counte i to count the numbe of eceived ymbol falling with thee phae angle ove a fame. Without any noie, thee thee counte value ae zeo fo /4-Shifted QPSK ignal and thee i ome value in each counte fo 8-PSK ignal. By computing the aveage value of thee counte, the modulation type can be identified. If C1 C2 C3 3 C1 C2 C3 8PSK 3 ShiftedQPSK 4, declae the modulation type a 8-PSK, ele the modulation type declaed a /4-Shifted QPSK. It hould be mentioned hee that ince /4-Shifted QPSK ignal ha 8 contellation point, the MSE citeion baed tet will claify it a 8-PSK ignal. The pocedue decibed above can ditinguih /4-Shifted QPSK fom 8-PSK baed on the elative phae diffeence. IV. EW ALGORITHM FOR RECOGITIO BETWEE 8-PSK AD /4-SHIFTED QPSK WITH SIMULATUI RESULTS In thi ection we popoe ou new method fo efficiently ditinguihing 8-PSK ignal fom /4-Shifted QPSK ignal. Fo clea compae with peviou pape eult, we fit imulate the algoithm popoed in [1]. To ditinguihing between 8-PSK and -Shifted QPSK ignal, we conideed 2000 ymbol pe fame of eceived ignal. Fig. 5 and Fig. 6 how the elative phae diffeence ditibution in an AWG and multi-path fading channel epectively. We could et a thehold on the aveage of occuence of two ignal. It i een that the two cuve ae not well epaated pecially at SR lowe than 15dB. Figue 6. Aveage of occuence of 8-PSK and /4-Shifted QPSK ignal v. SR in Multi-path fading channel. Fo bette detection of /4-Shifted QPSK ignal fom 8- PSK ignal it i neceay to mak both amplitude and phae vaietie of neighboing ymbol of eceived ignal. In pio method (among [1]), the emphai i only on phae diffeence of neighboing ymbol. Fo the ake of claity of new method, we hould attend on phae changing diagam of 8-PSK and /4-Shifted QPSK ignal a in Fig. 7. With defining the diffeence of two neighboing ymbol ignal a V ( i) ( i) ( i 1), we can depict each ample of the ignal V (i), a a vecto in ignal pace. Afte depicting all vecto in ignal pace, we find that the diffeence vecto illutation of /4-Shifted QPSK ignal i a ub-et of the diffeence vecto illutation of 8-PSK ignal (ee Fig.8). Figue 7. Phae changing diagam of (a) /4-Shifted QPSK, (b) 8-PSK Figue 5. Aveage of occuence of 8-PSK and /4-Shifted QPSK ignal v. SR in an AWG channel. (Recognition method ued in [1]) Figue 8. Phae diffeence vecto of neighboing ymbol fo (a) /4- Shifted QPSK and (b) 8-PSK ignal
So we et /4-Shifted QPSK point in ignal pace a contant in eceive libay and calculate the minimum Euclidean ditance between each eceived ymbol fom thee fixed point. Then we ave the mean of thee ditance in an accumulato fo each ignal. Finally we compae the value of accumulato and make the deciion; the one which ha minimum value i /4-Shifted QPSK ignal and the othe i 8- PSK ignal. With thi method, detection ability i well impoved. Fig. 9 and Fig. 10 how the imulation eult of thi method in AWG and multi-path fading channel epectively. To et a contant thehold we have un extenive imulation in vaiou condition of SR and vaiou numbe of eceived ignal ample. It i een that the two cuve ae well epaated in both channel except in vey low SR in multi-path fading channel. Alo, we mut note that thee eult ae achieved in mall numbe of eceived ignal ample which i elatively the ame a the cicumtance in eal opeation, whee we have limited numbe of ignal ample and little time to make coect detection. We have an extended compaion of ecognition accuacy of ou new method to ditinguihing /4-Shifted QPSK ignal fom 8-PSK ignal, againt the olde method [1, 8]. The eult have been hown in table 1 and table 2 fo AWG and multipath fading channel epectively. Fom table 1 and 2 it i obeved that in old method, ecognition accuacy depend on numbe of ample while in ou new method the ecognition accuacy appoximately become independent fom numbe of ample. Alo it i obeved that in new method, in the AWG channel and in SR above 7dB, the ecognition accuacy become independent of SR. Thi ate i SR above 15dB fo multi-path fading channel. Theefoe when knowledge about the SR i available, an adaptive cheme can be deigned to change the ample to attain highe accuacy. TABLE I. SR (db) 7 db 10dB 12 db o. of Sample 200 200 200 RECOGITIO ACCURACY I AWG CHAEL % Recognition Accuacy Modulation Type Old ew 67.25 77 8-PSK 73.87 77.25 /4-Shifted QPSK 78.5 88.87 8-PSK 81.62 89.12 /4-Shifted QPSK 86.25 93.62 8-PSK 90.12 94.5 /4-Shifted QPSK 91.75 99.87 8-PSK 94 100 /4-Shifted QPSK 99 100 8-PSK 99.37 100 /4-Shifted QPSK 99.87 100 8-PSK 100 100 /4-Shifted QPSK 99.12 100 8-PSK 99.5 100 /4-Shifted QPSK 100 100 8-PSK 100 100 /4-Shifted QPSK 100 100 8-PSK 100 100 /4-Shifted QPSK Figue 9. Mean of minimum Euclidean ditance of 8-PSk and /4-Shifted QPSK ignal v. SR in AWG channel. Figue 10. Mean of minimum Euclidean ditance of 8-PSk and /4-Shifted QPSK ignal v. SR in multi-path fading channel. TABLE II. SR (db) 10 db 15 db 20 db RECOGITIO ACCURACY I MULTI-PATH FADIG CHAEL o. of Sample 2000 2000 2000 % Recognition Accuacy Modulation Type Old ew 51.87 68.75 8-PSK 54.87 41.75 /4-Shifted QPSK 46.5 64.12 8-PSK 61.62 54.12 /4-Shifted QPSK 47.25 57.5 8-PSK 62.75 62.87 /4-Shifted QPSK 66 71.5 8-PSK 61.25 76.25 /4-Shifted QPSK 69 78.37 8-PSK 65.75 85.62 /4-Shifted QPSK 78.37 89.37 8-PSK 75.62 92.5 /4-Shifted QPSK 84.5 97.5 8-PSK 81.25 97.75 /4-Shifted QPSK 93.62 99.75 8-PSK 91.75 99.87 /4-Shifted QPSK 98.37 100 8-PSK 97.75 100 /4-Shifted QPSK
V. COCLUSIO In thi pape we have peented a novel appoach to automatic ecognizing and ditinguihing /4-Shifted QPSK ignal fom 8-PSK ignal. We alo eviewed automatic blind ecognition of diffeent QPSK ignal. Ability of blindly ditinguihing among the QPSK vaiant will have ueful application in SDR. The pefomance of the popoed cheme ha been teted in AWG and fading channel. Simulation eult how that the popoed cheme give vey good ecognition pefomance at low SR value in AWG a well a fading channel. The popoed cheme ha been found to be attactive both in tem of computational complexity and accuacy of ecognition. Alo it ha been obtained though imulation eult that at a fai SR, thi method could act independently fom SR and numbe of ample o the SDR ytem could et a contant thehold fo ecognition. REFERECES [1] M. Vatam aik, A. Mahata, R. Bhattachajee and H. B. emade, "Blind adaptive ecognition of diffeent QPSK modulated ignal fo oftwae defined adio application", Poceeding of the 1 t Intenational Conf. on Communication Sytem Softwae and Middlewae, Januay 2006. [2] C. S. Weave, C. A. Cole, R. B. Kumland and M. L. Mille, "The automatic claification of modulation type by Patten ecognition", Stanfod electonic laboatoie, Technical epot o.1829-2, 1969. [3] F. F. Liedtke, "Compute imulation of an automatic claification pocedue fo digitally modulated communication ignal with unknown paamete, "Signal Poceing, vol. 6, pp. 311-323, 1984. [4] E. E. Azzouz, A. K. andi, "Pocedue fo automatic ecognition of analog and digital modulation", IEE Poceeding on Communication, vol. 143, pp. 259-266, 1996. [5] B. G. Mobaei, "Digital modulation claification uing contellation hape", Signal Poceing, vol. 80, pp. 251-277, 2000. [6] S. S. Soliman and S. Z. Hue, "Signal claification uing tatitical moment, IEEE tanaction on communication, vol. 40, pp. 908-915, 1992. [7] Chung-Yu Huang, A. Polydoo, "Likelihood method fo MPSK modulation claification", IEEE Tanaction on communication, vol. 43, pp. 1493-1504, 1995). [8] M. Vatam aik, A. Mahata, R. Bhattachajee and H. B. emade, "An automatic blind ecognition algoithm fo MPSK ignal baed on MSE citeion", Poceeding of 2 nd Intenational Conf. on E- Buine and Telecommunication etwok, ICETE '05, Reading UK, Octobe 2005.