Model Developent for the Wideband Vehicle-to-vehicle.4 GHz Channel Guillero Acosta and Mary Ann Ingra School of ECE, Georgia Institute of Technology, Atlanta, GA 333-5, USA gte437k@ail.gatech.edu, ai@ece.gatech.edu Abstract Statistical channel odels are presented for a frequency selective vehicle-to-vehicle or obile-to-obile wireless counications link in an expressway environent in Atlanta, Georgia, where both vehicles travel in the sae direction. The odels were developed fro easureents taken using the direct sequence spread spectru (DSSS) technique at.45 GHz. The odels ply a non-separable channel with a persistent Rician behavior across ultiple odel taps. Key words: high obility channel, per-tap spectra, DSRC, wideband RF channel odeling, te-selective channel.. INTRODUCTION A standard has been developed for vehicle-tovehicle (VTV) high-speed data counications in the 5.9-GHz Intelligent Transportation Systes Radio Service (ITS-RS) Band. This dedicated short range counications (DSRC) standard [] defines a short to ediu range service, which supports both public safety and private operations in roadside-to-vehicle and VTV counication environents. Exaples of VTV applications include warnings for approaching eergency vehicles, pending intersection collisions, and road hazards. This paper applies the ethod of Mohr et al [] to extract statistical odels fro easureents obtained fro the expressway site of a VTV channel sounding capaign perfored at.4 GHz in support of the VTV part of the standards developent. Related works include theoretical -D [3] and 3-D [4] double-obility odels, flat-fading VTV easureents for the highway [5], and per-tap Doppler spectra for the fixed-to-vehicle channel [6]. Soe exaples of long and short ter per-tap Doppler spectra for the VTV channel were presented by the authors in [7]; to the author s best knowledge, these were the first reported per-tap VTV channel easured spectra. In this paper, we use an existing odel extraction approach to deterine an appropriate odel for the expressway channel. The authors gratefully acknowledge the support for this work provided by the National Science Foundation under Grant No. CCR-565, and by ARINC, Inc., Contract No. DTFH699-C-8.. MODEL DEVELOPMENT In this section, we start by describing the statistical procedure to obtain the paraeters describing the odel. We then present the obtained odels, and we evaluate the by coparing the synthesized odel obtained fro the statistical analysis with the actual recorded channel... Methodology The channel is easured using the spread spectru approach described in [3]. The recorded channel is sapled every τ = / B = 5 ns, where B is the easureent bandwidth of MHz. The first step in extracting the odel is to separate the fast fro the slow fading characteristics. We can obtain the slow fading by obtaining the local average power [7]. For this, we require uncorrelated saples of the pulse response (IR) over an appropriate distance that will average out the short-ter fading and will not sooth out the longter fading. This length is usually in the range of to 4 wavelengths, and a sapling at.8 λ is usually enough to achieve uncorrelated saples. We calculate the power delay profile (PDP) by averaging the agnitude squared of the IR over the distance Pn ( τ) = hnor ( n τ, k t) n N () k = where k t is the fixed observation instant of the k-th single IR, is the total nuber of IRs, and h nor indicates short-ter fading only. Next, we deterine the significant part of the PDP by discarding all parts that are ore than 35 db down fro the strongest path. Let L be the length of the significant part of the PDP divided by τ. Let M be the nuber of taps in the odel, such that M < L. The L saples are divided into M groups. If the result L/ M = I is not an integer, saples with zero aplitude are added to the end of the PDP. The tapped delay value in each group is found by []
+ I ( τ) n τ P n n= τ = for M () + I P n n= ( τ ) spectra, we can notice soe sall spikes on the sides of the spectra. These spikes were likely caused by transient paths reflecting fro highway overpasses. where i is the saple nuber at the start of each group. The delay of the first tap is τ =. Knowing the delay position of the tap, we can obtain the tap agnitude as follows: i + I h k t hnor n k t I ( τ, ) = ( τ, ). n= (3) So we get K coplex tap aplitude values for each tap with delay τ. The ean aplitude corresponding to each τ is the average value of the aplitude values. We deterine the Rice factor for each tap using the oent-ethod [9], [] as follows: K k= G = h ( τ, k t) h ( τ, k t) v k = k= k= = ( τ, ) v k = V h k t G Rice = k = V ( τ, ) h k t V Finally, we use the Welch algorith to estate the Doppler spectru per tap... Resulting Tap Characteristics For the expressway location, we recorded 4 tensecond takes. We divided each take into.7-second segents for their analysis. Please note that each take corresponds to a physically different te and location, i.e., each take could be inutes and kiloeters apart fro another. In Figure, we show the PDPs coputed for all the.7 s segents of the 4 takes. We observe that ost of the PDPs die out before µs, because of the generally confined nature of the channel. Assuing that ost coercial channel eulators are six or taps per channel, we decided to generate six and -tap channel odels. First, we generated a odel for each of the.7 s segents, and then, we generated overall odels. We present the 6 and -tap scattering functions for the overall odels in Figure and Figure 3, respectively. The corresponding statistical paraeters are given in Table. The first thing we notice in the Doppler spectra is the LOS-like or Rician behavior for each of the first paths. This suggests that there are any reflectors traveling at a silar speed inside a 45 locus of an ellipse defined by the reflecting path length. We also notice a widening of the Doppler spectra for the later taps. This widening is in accordance to the results of [5]. In several of these later. (4) Figure : Expressway PDPs Table : 6 and tap odel results Tap Delay, ns Magnitude, db K Rice 59 53.. 93.4.9 74 7 -. -7.6. 5.7 3 85 78-8.3-4.9.3 4.3 4 436 4 -.5-8.6.7 3.5 5 667 34-3.3 -...9 6 5 393-5.3-3.8.4.7 7 53-5.8.3 8 6-5.4.4 9 79-4.4. 84-4.3. 7-5.8. 55-7.6. Figure : 6 tap odel scattering function We can synthesize the channel using filtered noise on each tap, where the filter characteristics atch the spectra of the statistical odel []. As an exaple, the estated scattering function of the synthesized tap odel is shown in Figure 4, which represents a longte (i.e.,.7 s) estation. To confir the short-te behavior, we estated the long-ter and short-ter Doppler spectra of silar taps of the recorded and
synthesized channel. For the long-te spectru, we used the approxately 7, IRs in.7 s for the estate, and for the short-te spectra, we obtained the estate every,48 IRs or approxately every 5 ns. In Figure 5, we show the resulting Doppler spectra for the third tap of the synthesized odel. In Figure 6, we present the results of a latter tap of one.7-second segent of one of the takes of the recorded channel. As we can see in these figures, the short-te spectra of the synthesized channel have transient behavior silar to that of the real channel. - -4-6 -6-4 - 4 6 Sulated Channel Path 3 Long Te Power Spectru - -4 Sulated Channel Path3 Short Te Power Spectra -6 - -5 5 Figure 5: Sulated short te Doppler spectra and long te Doppler spectru for the third tap of the tap odel scattering function Figure 3: tap odel scattering function - - -3-4 -5 Bin Short Te Power Spectra -6-6 -4-4 6 Bin Long Te Power Spectru - - -3-4 -5-6 - -8-6 -4-4 6 8 Figure 6: Measured short te Doppler spectra and long te Doppler spectru of a latter tap of one of the resulting scattering functions Figure 4: Synthesized tap odel scattering function 3. Testing Using a DSRC Sulink Model To validate the statistical channel odel, we copared the bit error rate (BER) perforance of a DSRC link over the recorded channel and over the synthesized channel. We noralized all the channels so that their PDPs had unit area; this includes separate noralization for each easured segent. We developed a Sulink odel that follows faithfully the specifications of the DSRC standard [] shown in Table. The odel includes all the transission odes and coding. We designed the receiver to include frequency offset copensation based on the algorith described in [], and channel adaptive equalization using the pilot tones following the specifications in [3]. The odel also contains a signal-to-noise ratio (SNR) threshold detector to adaptively change aong the eight transission odes. The noise is defined as the ean square of the difference between the received coplex sybol and its corresponding constellation point. The OFDM sybol is odulated in only 5 subcarriers out of 64 and has 8. µs duration with a.6 µs guard band 3
interval. The total occupied bandwidth is 8.3 MHz. No theral noise was added in our sulations. Data Rate, Mbps Table : DSRC standard specifications Modulation Coding Rate Coded Bits per OFDM Sybol SNR Threshold db 3 BPSK / 48 < 4.5 BPSK 3/4 48 6 PSK / 96 9 PSK 3/4 96 4 6-AM / 9 8 8 6-AM 3/4 9 4 64-AM /3 88 6 7 64-AM 3/4 88 8 3.. Results For the odel validation procedure, we perfored five different experents or processes: recorded channel, tap overall odel, tap odel for each.7 s segent, 6 tap overall odel, and 6 tap odel for each.7 s segent. We atched the length of these projects to that of the recorded channel. For the recorded channel, we have 68 (4 takes x files).7 s segents that we further divided in.3 s segents (because of eory litations) for a total of 336 runs or sulations. Each.3 s segent contained, IRs. For the sulation, we used 68 OFDM sybol fraes, four of which were training sybols. Each.3 s segent contained 564 fraes, and the total nuber of fraes per process was 89,37. For the overall 6 and tap odel processes, we used the scattering functions of Figure and Figure 3 as the filter teplates for 68.7 s different rando sequences that produced 4,3, different IRs for each process. For the 6 and tap per segent processes, we obtained a scattering function for each.7 s segents, producing 68 different odels. We then used these 336 to filter another set of 336.7 s rando sequences. The adaptive odulation syste of the odel worked in a frae by frae basis. We recorded the BER, SNR, bit rate, and ode for each frae of each process. In Figure 7 and Figure 8, we show the BER histogra for each of the five processes in ters of.3 s segents and in a frae by frae basis. In Figure 9, we present the results for the bit rate in a frae by frae easure. This last figure provides the sae inforation as if we were to present the results using either the SNR or ode. By just looking at the BER results, you ight conclude that the tap per segent odel is the closest to the recorded channel, but by looking at the bit rate results, we see that the closest is the tap odel. We can obtain a better conclusion by easuring the overall BER (i.e., total errors divided by total nuber of bits transitted in the 89,37 fraes) for each process as shown in Table 3. Here we can see that the best atch is the tap odel. Another portant result of the channel sounding capaign is the worst case channel. In Figure, we present the PDP of the.3 s segent with the highest BER of.56. We also provide its scattering function in Figure. We observe that there are soe spikes in the PDP corresponding to excess delays of close to µs and beyond 3 µs. Even though the heights of these spikes are ore than 3 db down fro the axu value of the PDP, they correspond to excess delay values greater than.6 µs (the length of the cyclic prefix), and therefore they introduce intersybol interference, which is particularly harful to an OFDM receiver. We can also notice a faster widening of the Doppler spectra with any spikes throughout the spectral bandwidth. Figure 7: Histogra of the bit error rate for each processed.3 s segent Figure 8: Histogra of the bit error rate for each processed frae 4
Figure 9: Histogra of the bit rate of each processed frae Table 3: Final bit error rate for each process Bit Error Rate Recorded Channel 9.76e-3 6 Tap Overall Model 6.83e-3 6 Tap Model for Each.7 s Segent 4.6e-3 Tap Overall Model 9.56e-3 Tap Model for Each.7 s Segent 5.3e-3 fractional power, db Worst Case (BER =.56) PDP 5-5 - -5 - -5-3 -35-4 -45 3 4 5 6 delay, µs Figure : Worst case PDP for a BER =.56 Figure : Worst case (BER =.56) scattering function 4. CONCLUSIONS A new tap statistical odel has been developed for the VV frequency selective highway channel when both vehicles travel in the sae direction. It has the novel feature of a significant Rician coponent in first taps. The odel was validated in ters of a BER easured copared to the actual recorded channel using a coplete DSRC sulation tool. REFERENCES [] ASTM E3-3, Standard Specification for Telecounications and Inforation Exchange Between Roadside and Vehicle Systes 5 GHz Band Dedicated Short Range Counications (DSRC) Mediu Access Control (MAC) and Physical Layer (PHY) Specifications, ASTM International, www.ast.org, [] W. Mohr, Modeling of wideband obile radio channels based on propagation easureents, in Proc. 6 th Int. Syp. Personal, Indoor, Mobile Radio Counications, vol., pp. 397-4, 995. [3] C. S. Patel, G. L. Stüber, and T. G. Pratt, Sulation of Rayleigh faded obile-to-obile counication channels, in Proc. of IEEE Vehicular Technology Conf., vol., pp. 63-67, October 3. [4] F. Vatalaro and A. Forcella, Doppler spectru in obile-to-obile counications in the presence of three-densional ultipath scattering, IEEE Trans. on Vehic. Tech., vol. 46, no., pp. 3-9, 997. [5] J. Maurer, T. Fügen and W. Wiesbeck, Narrow-band easureent and analysis of the inter-vehicle transission channel at 5. GHz, in Proc. of IEEE Vehicular Technology Conf., vol. 3, pp. 74-78,. [6] X. Zhao, J. Kivinen, P. Vainikainen, and K. Skog, Characterization of Doppler spectra for obile counications at 5.3 GHz, IEEE Trans. Vehicular Technology, vol. 5, no., pp. 4-3, 3. [7] G. Acosta, K. Tokuda, and M. A. Ingra, Measured joint Doppler-delay power profiles for vehicle-to-vehicle counications at.4 GHz, in Proc. of IEEE Global Teleco. Conf., vol. 6, pp. 383-387, 4. [8] W. C. Y. Lee, Estate of local average power of obile radio signal, in IEEE Transactions on Vehicular Technology, vol. 34, No., pp. -7, February 985. [9] A. Abdi, C. Tepedelenlioglu, M. Kaveh, and G. Giannakis, On the estation of the K paraeter for the Rice fading distribution, in IEEE Counications Letters, vol. 5, No. 3, pp. 9-94, March. [] L. J. Greenstein, D. G. Michelson, and V. Erceg, Moent-ethod estation of the Ricean K-factor, in IEEE Counications Letters, vol. 3, No. 6, pp. 75-76, June 999. [] M. C. Jeruch, P. Balaban, and K. S. Shanugan, Sulation of Counication Systes: Modeling, Methodology, and Techniques, Second Edition, Kluwer Acadeic Press, Boston. [] J. J. van de Beek, M. Sandell, and P. O. Borjesson, ML estation of te and frequency offset in OFDM systes, in IEEE Transactions on Signal Processing, vol. 45, pp. 8-85, July 997 [3] M. H. Hsieh and C. H. Wei, Channel estation for OFDM systes based on cob-type pilot arrangeent in Foratted: Bullets and Nubering Foratted: Bullets and Nubering 5
frequency selective channels, in IEEE Transactions on Consuer Electronics, vol. 44, No., pp. 7-5, February 998. 6