Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 Perforance of Antenna Variable Modulation for urbo MIMO ransission in Frequency-Selective Channels * Christian Schneider, Marcus Großann and Reiner hoä Electronic Measureent Research Lab echnische Universität Ilenau PF 1565, 98684 Ilenau, Gerany christian.schneider@tu-ilenau.de Abstract In this contribution the robustness and reliability of a broadband single carrier MIMO transission in significantly changing channels is under investigation. Real field easureent data is used to show the effectiveness and influence of the proposed MIMO lin adaptation schee, where for different transit antennas variable signal odulations are considered. Furtherore the perforance of two different concepts, the ulti level bit interleaved coded odulation () and the coon BICM, for urbo MIMO Equalization with variable higher odulations is copared. It was found, that the teporal and spatial ultipath structure is strongly influencing the robustness of the MIMO transission. An adaptation of the signal odulation per antenna iproves the reliability on the signal detection. Keywords turbo io equalization, per-antenna-rate-control I. INRODUCION Multiple-input ultiple-output (MIMO) air interfaces based on antenna arrays at both the transitter as well as the receiver side are shown to be a ey technology for obile internet and next generation wireless systes. he requested high bit rate of these wireless lins can only be reached with broadband MIMO systes. A low-coplexity signal separation ethod which jointly exploits the spatial as well as the teporal structure of frequency selective MIMO channels was proposed in [1] the urbo MIMO Equalizer (ME). ere for frequency-selective MIMO channels the coputational coplexity of the iniu ean square error (MMSE) based equalizer grows only in cubic order with the nuber of parallel streas/users and their channel eory lengths, whereby for a axiu lielihood sequence estiation (MLSE) or axiu a-posteriori (MAP) detection the coplexity increases exponentially. Recent research results [4] [5] have shown that the perforance of ME schees strongly depend on the ultipath propagation characteristic. o provide a MIMO counication which is robust and reliable under changing propagation condition an adaptive optiization regarding the antenna configuration (nuber of parallel transissions as well as antenna selectio, diversity concepts, channel coding, sybol odulation, etc. of the signal transission over changing MIMO channels is required. In this publication we focus on antenna variable odulation for transissions based on the ME. wo nown concepts for higher odulation and detection based on the turbo equalization [] [3] are here extended to the MIMO case and to a syste concept with variable odulation for different transit antennas. he perforances are assessed and copared using realtie MIMO easureent data. Channel sounding techniques provide the possibility to evaluate the perforance of radio ultiple access and signal processing schees under realistic propagation conditions. Coplex channel ipulse responses (CIR) gathered through MIMO easureents in a real field scenario with different propagation characteristics have been used in lin-level siulations. he identification of the ultipath channel by eans of high resolution paraeter estiation is essential for investigating the influence of the spatial and teporal structure on the perforance of the ME. herefore this paper eploys the results of double-directional channel sounding experients for MIMO lin level siulations [6]. II. adaptive flow control ADAPIVE SPACE IME SIGNAL RANSMISSION A. ransitter/receiver Configuration he considered syste is based on the ME [1], whereby two different concepts for the generation and iterative detection of higher signal constellations were found in [] [3] and extended to MIMO counication. he channel encoding and odulation schee for each transit antenna is in the first case [] the ultilevel bit-interleaved coded trans. odul trans. odul lin adaptation MIMO Channel feedbac lin MIMO Equalizer - channel estiation rec. odul rec. odul lin adaptation adaptive flow control Figure 1: Adaptive broadband MIMO syste based on ME with and BICM * his wor was partly supported by the Geran Federal Ministry of Education, Science Research and echnology under the yeff project line.
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 ENC,1 DEMUX MAPPER ENC MAPPER ENC,M (a) (a) -1-1 DEC,1 DEC,M MUX -1 DEC Figure : (a) transit odule, (b) receive odule (b) odulation () with square QAM sybol apping and in the second case [3] the bit-interleaved coded odulation (BICM). Furtherore the MIMO syste concept relying on ME with and BICM were extended to support different signal odulations for each transit antenna. In Fig. 1 a siplified syste concept is shown. he transitter consists of an adaptive flow control and transit odules, each for every antenna. An adaptive flow control is assued to schedule the inforation bits in N data streas, regarding the selected odulation. he and BICM blocs in Fig. 1 include the signal processing on the transit as well as the receive side. he nuber of receive blocs are equal to the nuber of transit blocs and a receive bloc is always associated to exact one transit bloc. he transission ode of a transit bloc is deterined through the receiver via a feedbac channel. he transitted ultilevel coded signal of the -th transit bloc (Fig. a) is constructed as follows: At first, the inforation bits are deultiplexed into M parallel levels and convolutionally encoded. Next, the encoded bits are interleaved with different rando interleavers and then BPSKodulated. he resulting sybols of the different levels are grouped into sequences of M bits. he apper can be seen as a linear cobiner of the M BPSK-odulated sybols, so that each bit in a group is ultiplied with a level-specific coplex weighting factor z and sued up into a QAM sybol. he coplex weighting factor z is real for odd and iaginary for even. Each transitted QAM sybol is then given by s ( = z c(, (1) where z = [ z 1,..., z,..., zm ] is the strea specific appingvector for antenna with its levels being ordered by decreasing aplitude, and 1 c = [ c,..., c,..., c ] is the group of BPSK-odulated bits fro the individual coding levels. he coplex signal points of the apper are fro the set = { s1,.., s M }. Notice that the signal sets S S M Figure 3: (a) BICM transit odule, (b) BICM receive odule (b) with = 1.. N can be different. he corresponding signal constellation consists of super positioned 4-QAM constellations []. For exaple, the apping vector of a 16 QAM constellation (two super positioned 4-QAM constellations) is given as z = / 5[1, j,.5,.5 j]. For a bitinterleaved encoding and odulation schee (Fig 3a) the inforation bits of the -th strea are encoded by only one convolutionally encoder and interleaved using a rando perutation function. Each transit sybol s ( is M generated by apping M interleaved code bits to a -ary signal constellation S. In this paper only 4 QAM and 16 QAM signal constellations with gray-apping are considered. hus, the apping in the BICM case is non-linear. In a turbo transission syste the ultiple coded levels of a schee can be decoded in parallel without interdecoder inforation exchange []. ence, a receive bloc (Fig. b) consists of M channel decoders, which wor in parallel and in each case are separated by interleaving and de-interleaving. A channel decoder is always associated to exact one level of the proper transit odule. Because of the linear apping a coplex sybol-to-bit soft de-apping generally required in equalization of high order odulation can be avoided. hose are needed in the BICM for turbo equalization [3]. In Fig. 3b a BICM receive bloc is shown. It consists of only one channel decoder separated by an interleaver and de-interleaver. B. Signal Modell he data sybols s ( are transitted across a frequency-selective channel, which is assued static over a frae and perfectly nown at the receiver. he radio channel between each pair of M receive and N transit antennas is odeled by coplex finite ipulse responses h, ( l) with L taps. he receive signal at antenna can be represented as L 1 N r h ( l) s ( n l) + σ v = l = = 1, n ( ), ()
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 where v ( is an additive white Gaussian noise saple with which is diagonal, since the independence between coded bits variance 1. Furtherore, a copact atrix-vector notation of is assued [3]. For an efficient coputation of (7) the atrix () can be written in the for inversion lea [1] is applied. Finally, the output of the MMSE filter can be coputed as r = s( + σ v(, (3) by introducing a space-tie MIMO channel atrix and sˆ = α [ rˆ( + s h ], (11) receive and transit space-tie saple vectors r ( and with s (, siilar to [1]. Additionally, the vectors h are defined as h = h 1 ( L 1)... hm ( L 1)... h1 ()... hm ()], which are the N central coluns of the atrix. III. URBO MIMO EQUALIZAION FOR PER ANENNA VARIABLE IGER MODULAIONS A. Coon Approach for urbo MIMO Equalization he urbo MIMO equalizer in this paper is based on the algorith presented in [1], extended for iterative detection of higher signal constellations. he equalizer odule perfors soft-interference cancellation for each transit sybol, utilizing a-priori sybol probabilities. On the basis of the available a-priori inforation, the first and second order statistics { s } = s jpa ( s ( n = s j ) s = ˆ E ) (4) s j S { s } ˆ E{ s } s { s } = s j Pa ( s ( n = s j ) var = (5) E ) s j S are coputed for each transitted sybol s ( with = 1..N and n = 1.. B. he sybol a-priori probability P a is obtained fro the corresponding bit a-priori log-lielihood ratios (LLRs), produced by the channel decoders [3]. he residual signal of the soft-cancellation can be specified with (3) and (4) as rˆ = r( s(, (6) where the vector s ( coprises all a-priori sybol eans. For further filtering of the residual signal, an instantaneous MMSE filter ( is applied. o copute the filter taps (, the MMSE criterion arg in E s (ˆ( r + h s ) (7) is solved for each. he solution to the MMSE optiization is shown to be 1 [ Σ( + (1 var{ s }) hh ] h =, (8) where Σ ( is the covariance atrix of the residual signal in (6) and is given as Σ( n ) = Λ( + σ I. (9) ere, I is the identity atrix of size (LM ) and Λ ( the covariance atrix { var( s( ))} Λ = diag n, (1) ( 1 + ( 1 var{ s }) h ) 1 α = (1) 1 = h Σ. (13) he corresponding sybol extrinsic probabilities are then approxiated as Gaussian PDFs p ( sˆ s = si ) with si S using an equivalent Gaussian channel assuption at the output of the MMSE filter [1]. Based on the sybol extrinsic probabilities and the a-priori inforation, the bit extrinsic LLR can be coputed for each encoded bit [3]. B. Equalization For a schee with square QAM sybol apping, an efficient turbo equalizer algorith can be defined. he apper can be seen as a linear cobiner of BPSK-odulated sybols. hus, the signal constellation is generated through binary sub-constellations fro the individual coding levels and the algorith can consider each single level as the desired signal. he required equalizer can be obtained by utilizing the algorith described in the previous section for BPSK odulation, integrating the level wise odulation with the weight z. Due to the binary odulation, the first and second order statistics can be indicated for each level bit c ( with = 1..M as d λ, ( c = tanh (14) d, ( ) var{ ( )} 1 tanh λ n c n = (15) d where λ, is the bit a-priori LLR, which is generated by decoder (, ). he residual signal in (6) can now be written as rˆ = r( Zc(, (16) where Z is a bloc-diagonal atrix, which contains the specific apping vectors of all streas and is given as z1 L L O M Z = diag M z M (17) M O L L z N and c ( is the vector, which coprises all bit a-priori sybol eans. he covariance atrix of the residual signal (9) becoes Σ( n ) = ZΛ( Z + σ I (18)
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 with the atrix of equation (1) diag{ var( c( ) } Λ =. he MMSE filter output equations (11)-(13) can therefore be rewritten as with [ rˆ( c ( n h z ] cˆ = α, + ) (19) α = 1+ c, h z () 1 *, = h Σ z. (1) Finally, the bit extrinsic LLR for the M parallel channel decoders are coputed for each = 1.. N and = 1..M as [] with { cˆ } 1 e R λ, = 4 () 1 µ,, = α, µ h z. (3) C. BICM Equalization he equalizer algorith for a BICM schee was presented in section A. For the coputation of the bit extrinsic LLR, the sybol extrinsic probabilities are required [11]. hose are calculated as ( ) 1 sˆ µ si p sˆ s = si = exp (4) ν π ν for each s S. he statistics µ ( and ν are given by i = α µ µ µ h (5) ν =. (6) Based on the sybol extrinsic probabilities in (4) and the a- priori sybol probabilities, the bit extrinsic LLR for the -th channel decoder is coputed as [1] e, p ( sˆ s = s ) exp( L ( s )) i a i 1 si S, p i a i ) si S, λ = ln, (7) where 1 S, and ( sˆ s = s ) exp( L ( s ) S, define the subsets of S where the bit c () taes the value 1 and, respectively. he a-priori sybol probability L a ( s i ) for sybol si S in (7) is based on all bit a-priori LLRs, except for that particular bit. N s tat. dyn. Figure 4: Measureent Scenario X 1 X x RX Ilenau, is copletely enclosed by a building of about 15 height. Furtherore several various etal objects (container, esh fence and tubes) were found within the courtyard. An 8 eleent unifor linear antenna (ULA) with separate ports for horizontal and vertical polarization was utilized on the receive side, whereby the antenna was ounted at a height of 1.67 and only the vertical polarization was easured. On the opposite a unifor circular array consisting of 16 onidirectional eleents was fastened at a height of.1 and oved at waling speed fro position X 1 to X. All easureents have been perfored at 5. Gz carrier frequency and with a bandwidth of 1 Mz. he easureent trac can be separated into three parts, where different ultipath characteristics were found. ere the spatial channel properties are highlighted in ters of root ean square (RMS) Rx and x aziuth spreads (direction of arrival and direction of departure), see Fig. 5. A continuative identification of ultiple path coponents can be found in [5]. In the beginning of the easureent the transitter was not RMS Rx aziuth spread [ ] 4 3 1 N stat. dyn. x y 1 1 8 6 4 RMS x aziuth spread [ ] IV. MEASUREMEN BASED REALISIC SIMULAIONS Rx A. Measureent Scenario and Propagation Charaterization he MIMO easureent scenario under consideration can be classified as icro cell outdoor scenario for a wireless local area networ (WLAN) application with low obility. A setch of the scenario is shown in Fig. 4. he place, a large courtyard at the capus of the echnische Universität 1 4 6 8 18 Figure 5: RMS x and Rx aziuth spread along the entire easureent trac
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 oving, at this point a rich ultipath diversity with RMS x spreads around 1 were found. Starting fro X 1 the transitter oved under N (Non-Line-Of-Sight) condition for approx. 3. In the ean the x spreads vary around 6-7, which indicates ediu ultipath diversity. Whereby the Rx spreads for both N parts (stat. and dyn.) are considerable lower (ean of 15 ) due to the liited view of the ULA (+/- 6 regarding to the antenna loo directio. For the rest of the trac transitter and receiver had line of sight (), here naturally the spread values are significantly lower copared to the N cases. hese three different propagation characteristics (N stationary, N dynaic and dynaic) are considered in the subsequent perforance evaluation section. B. MIMO Syste Siulation Assuptions For sipleness - a MIMO syste based on two transit and two receive antennas was considered for all lin level siulations. Both, the BICM and concept use convolutionally encoding (C l =3 and R c =.5) and rando bloc interleaver, independent for each antenna as well as for each ultilevel. Every transitted frae consists of 18 sybols. According to the assuptions fro [4] the easureent data were selected and preprocessed for the target sybol rate of Msy/s per antenna. Whereby for the pulse shaping a root-raised cosine filter (roll-off=.5) was eployed. he resulting channel ipulse responses are odeled, corresponding to the delay coponents observed within the easureent data, using 4 delay taps. Only equipped with L = 9 teporal taps the receiver perfored the signal separation. his is sufficient to capture the significant part of the received energy within the channel ipulse responses. In principle the proble of transit power and signal to noise ratio (SNR) definition can be seen fro different viewing angles, in particular if the transition between N and for MIMO syste evaluation is considered. In this paper the goal is to analyze the perforance ipact of the changing ultipath structure, consequently a constant transit power and subsequently path loss influences are not in the focus here. noralized x Power [db] 35 3 5 15 1 5-5 Ant. 1 Ant. 1 3 4 5 6 7 8 9 1 Figure 6: Nor. transit power for sae odulation at both antennas herefore the signal to noise ratio (SNR) at the receiver is held constant and identical for each transitted inforation bit by an adaptive power control. Fig. 6 shows the resulting noralized transit power per antenna, whereby at both antennas the sae odulation is assued. With adaptive odulation per transit antenna different sybol to noise ratios according to the aforeentioned understanding would result. Furtherore the total transitted power is independent on the nubers of transit antennas. Considering this assuptions we are able to copare in a transparent way the lin perforance in ters of bit error rates () for MIMO systes with variable odulations per antenna. V. RESULS OF E REALISIC MIMO SIMULAIONS he general question within this paper is to analyze if two parallel transitted signals at higher odulation can be continuatively separated in channels with changing ultipath characteristics and which adaptation on the transit signal processing is necessary to reach a predefined degree of robustness. ere no specific paraeter is considered for adaptive control of the odulation schee, hence the focus is on its potential ipact. he signal odulation can be changed independently for each antenna between QAM and 16-QAM (both with gray apping). he axiu ean perforance of at an SNR of 9 db for the x ME systes was selected to be the threshold of robustness. 1 BICM x1: 16-QAM x: 16-QAM 1 x1: 16-QAM x: 16-QAM 1-1 - 4 6 8 1 1 14 16 18 Figure 7: curves for / ME based on BICM with 16-QAM under different propagation conditions 4 6 8 1 1 14 16 Figure 8: curves for / ME based on with 16-QAM under different propagation conditions
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 1 BICM x1: 16-QAM x: 16-QAM 1-4 6 8 1 1 BICM x1: QAM x: 16-QAM 1-4 6 8 1 Figure 9: along entire trac of / ME based on BICM considering different odulation setups per antenna In Fig. 7 and Fig. 8 the s after the 6 th iteration versus the SNR for ME with BICM respectively are shown for the different parts of the easureent tracs, see Fig. 5. Obviously for both higher odulation concepts the perforance strongly depends on the ultipath diversity within the channel, whereby the shows better robustness by lower ultipath characteristics copared to the BICM case. It is clear that both concepts can easily separate the two 16-QAM signals only under the propagation condition, for the other parts the threshold can not be reached. A ore detailed view to the ME s perforance is given in Fig. 9 and 1, here the position variant s are presented. For each position (quasi static channel realizatio 5 fraes with an independent noise assuption are transitted to ensure reliable results. Fig. 9 shows the at 9 db SNR and after 6 iterations of the ME syste with BICM. ere in a first configuration both antennas transit signals with 16-QAM (above) and in the second x1 with QAM and x with 16-QAM (below). For the sae 1 BICM 1 - x1: QAM x: 16-QAM x1: 16-QAM x: QAM 4 6 8 1 1 14 Figure 11: vs. SNR for / ME based on BICM with variable odulation (16-QAM and QAM) on different antennas 1 x1: 16-QAM x: 16-QAM 1-4 6 8 1 1 x1: QAM x: 16-QAM 1-4 6 8 1 Figure 1: along entire trac of / ME based on considering different odulation setups per antenna configuration Fig. 1 highlights the results. It was found that the robustness of BICM and based ME systes are significantly increased if the odulation schee only at one antenna is reduced. With QAM at both transit antennas the robustness would further increase, the was found to be below, but on the other side capacity and throughput are useless wasted the systes are not optiized. he MIMO receiver is not able to detect the two 16-QAM signals sufficiently regarding the and SNR constraint with decreasing RMS x aziuth spreads ( and further dyn.). ence a variable odulation per antenna is introduced to overcoe this proble. Fig. 11 ephasized the question: on which antenna which odulation should be selected? For the considered syste setup two constellations are possible, whereby only inor perforance differences occur even within the varying channel properties. In the following the setup with the lowest s is considered. he perforance of the adaptive odulation selection per transit antenna along the entire easureent trac is highlighted in Fig. 1. Furtherore the total data rate R regarding the adaptive MIMO syste rate is shown. For the syste concept reliable and robust broadband MIMO transission can be gained if the adaptive signal transission following the actual channel condition is considered the SNR and threshold to ensure robust counication can be reached. he BICM case shows slightly better perfors for the N parts, but for the concept can not fulfill the robustness constraints. ere a further adaptation step sees to be required. Fig. 13 suarized the results in ters of ean versus SNR for BICM and with adaptive odulation per antenna for the three different ultipath propagation conditions. Again the result shows that the with adaptive odulation at one antenna (fro 16- QAM to QAM) can reduce the erroneous transission and hence constant lin connection is possible. he curves for the and cases of are siilar to the of BICM and therefore are not clearly visible.
Proceedings IG Worshop on Sart Antennas, Munich, Gerany, 18th-19th of March 4 BICM variable M-QAM per antenna 1 dyn. 1-4 6 8 1 1 variable M-QAM per antenna dyn. 1 - R=8 Mbit/s R=6 Mbit/s R=6Mbit/s R=8 Mbit/s R=6 Mbit/s R=6Mbit/s 4 6 8 1 Figure 1: and data rate along entire trac of / ME based on BICM and with variable signal odulation per antenna CONCLUSION & OULOOK he research results based on easureent data highlighted the fact, that the perforance of spatial ultiplexing MIMO systes in particular with higher odulation strongly dependent on the ultipath characteristics within the channel. Furtherore the variable odulation per antenna can ensure the robustness and reliability of the MIMO transission and hence is essential to be used if the channel properties change even under N condition. Within propagation scenarios characterized by low ultipath diversity the shows perforance advantages copared to the coon BICM approach. MIMO easureent trials cobined with high-resolution channel characterization provide detailed insights into the physical propagation conditions in different real world scenarios. Using these ethods the perforance of the candidate MIMO concepts can be realistically evaluated and interpreted. In continuation of the research within this paper the derivation of proper control paraeter (e.g. spatial correlatio for lin adaptation will play an iportant role. By using a feedbac lin the transitter can be controlled by the receiver based on the channel/lin properties. Cobined with this also sart and low coplexity adaptation algoriths/strategies are necessary. Moreover the adaptive coding and antenna selection jointly with odulation changes is seen to be a hot topic for research. Future wor will also cover the optiization and enhanceent of MIMO algoriths for the transitter as well as for the receiver with respect to application and deployent issues. hese will also include scalability in ters of parallel users/data streas for the spatial ultiplexing MIMO concepts. In particular in channels with strong coponents and/or low ultipath diversity it could play an iportant role for perforance optiization. 1 BICM & 1 - BICM 4 6 8 1 1 14 Figure 13: : vs. SNR for / ME based on BICM & with variable odulation (x1: QAM/16-QAM and x:16-qam) ACKNOWLEDGMEN he authors wish to than MEDAV Gb for supporting their RUSK MIMO channel sounder and preprocessing tools for realistic siulations, furtherore the colleagues at the echnische Universität Ilenau for perforing the easureents and supporting the data analysis. REFERENCES [1]. Abe and. Matsuoto, Space-ie urbo Equalization in Frequency-Selective MIMO Channels, IEEE rans. Veh. echnol., vol. 5, no. 3, pp. 469-475, May 3 [] K. Kansanen and. Matsuoto, "urbo Equalization of Multilevel Coded QAM", Conf. Rec. SPAWC3, June 3, Roe, Italy [3] A. Dejonghe, L. Vandendorpe, urbo equalization for ultilevel odulation: a low coplexity approach, ICC, Conf. Rec. pp. 1863-1867, April 8 May,, New Yor [4] U. rautwein,. Matsuoto, C. Schneider, R. hoä, Exploring the Perforance of urbo MIMO Equalization in Real Field Scenarios, to be presented at the WPMC, onolulu, awaii [5] C. Schneider, R.S. hoä; U. rautwein,. Matsuoto, he Dependency of urbo MIMO Equalizer Perforance on the Spatial and eporal Multipath Channel Structure, IEEE VC-Spring, April 1-4, Jeju, Korea [6] R.S. hoä, D. apice, M. Landann, G. Soerorn, A. Richter, MIMO Measureent for Double-Directional Channel Modeling, IEEE echnical Seinar on "MIMO Counication Systes", London, Deceber 1 [7] http://www.channelsounder.de [8] R. S. hoä, D. apice, A. Richter, G. Soerorn, U. rautwein, MIMO Vector Channel Sounder Measureent for Sart Antenna Syste Evaluation, European rans. on eleco., E Vol. 1, No. 5, Sept./Oct. 1 [9] X.Wang and. V. Poor, Iterative (turbo) soft interference cancellation and decoding for coded CDMA, IEEE rans. Coun., vol. 47, no. 7, pp. 146 161, July 1999. [1] M. üchler, A. C. Singer, and R. Koetter, Miniu ean squared error equalisation using a priori inforation, IEEE rans. Signal Processing, vol. 5, no. 3, pp. 673 683, Mar.. [11] G. Caire, G. aricco, and E. Biglieri, Bit-interleaved coded odulation, IEEE rans. Infor. heory, vol. 44, pp. 97 946, May 1998. [1] S. Brin, J. Speidel, R. Yan, Iterative Deapping and Decoding for Multilevel Modulation, IEEE Global elecoun. Conf., Sydney, Australia, Vol. 1, pp. 579-584., Nov. 8-1 1998