Академический Документы
Профессиональный Документы
Культура Документы
Abstract— This paper attempts to compare the system down- The outdoor restaurant.
link performance of a smart antenna scheme for OFDMA that
utilizes channel covariance information with an opportunistic
MRS
multi-antenna scheme for OFDMA that uses fast limited feed-
back of the instantaneous channel condition. The schemes are
evaluated by means of simulations in a low load scenario and
In car
also compared to a simple HSDPA-like scheme. The smart Outdoors
8
the supportable rate for a small number of their strongest
beams. Based on this information, the base station schedules
6
users according to a version of the proportional fair algorithm
that updates the user rates after the allocation of all clusters
4
and beams (the "third" scheme in [13]).
In order to reduce the feedback load, the beamforming
2
scheme does not work on a sub-carrier level, but on clusters of
adjacent sub-carriers. Furthermore, only the strongest beams 0
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
across the clusters are fed back [6]. The relays are set up to Average number of users per sector
feed back the supportable rates on many beams if there are few
relays in the sector and fewer beams if there are many relays Fig. 3. System throughput versus load for the opportunistic scheme, Nr = 4.
in the sector. The total feedback rate per sector is therefore
relatively constant and at most around 100 kbps. Note that
the feedback rate was designed for mobile speeds around 100 D. Assumptions for HSDPA
km/h. Since the simple HSDPA scheme uses single-carrier com-
munication, the OFDM parameters in Table I are not valid.
TABLE I The chip-rate is 3.84 MHz. Spreading factors (SF) between 1
S IMULATION PARAMETERS . and 64 are used together with adaptive modulation. For the
Total bandwidth 3.84 MHz HSDPA system, the time-slot duration is 2.3 ms. Assuming
Number of sub-carriers 128 that only three bits per time-slot have to be fed back by
Cluster-size 4 sub-carriers the users (’keep modulation/SF’,’higher modulation/SF’ and
Sub-carrier spacing 30 kHz ’lower modulation/SF’), the feedback rate is only 1.3 kbps
Cyclic prefix length 5µs
Total OFDM symbol period 33.33+5=38.33 µs
per user.
Carrier frequency 2 GHz IV. S IMULATION R ESULTS
Time-slot length 32 OFDM symbols
Transmit power per sub-carrier The results obtained in the case of 4 antennas at each
and sector (P) 312.5 mW relay are illustrated in Figure 2–5. Results for more cases are
Noise power per sub-carrier summarized in the cost analysis in Section V.
and antenna (σz2 ) 2.38 · 10−16 W Figures 2 and 3 show the system throughput (sum rate per
Site-to-site distance 3km sector) as a function of the load (average number of users per
Relay speed 3km/h
Modulation OFDM, BPSK, M-QAM
sector) for the different solutions. For the opportunistic SD-
M ∈ {4, 16, 64, 256} OFDMA in Figure 3, using only one transmit antenna gives
the highest throughput. The reason is that an opportunistic
1
1
0.9
0.9
0.8
Cumulative Density Function
Fig. 4. CDF of the average user rate for the smart antenna scheme (adaptive
modulation), Nr = 4. Fig. 5. CDF of the average user rate for the opportunistic scheme, Nr = 4.
system needs many users to work well. Additional antennas more than one user accesses a relay at any time and thus the
(and beams) increase the interference significantly. For the low probability, pactive , of a relay being active is given by pactive =
loads studied here, the spatial multiplexing gain is too low to pUE-active NUE-per-relay . The mean number of active relays is given
compensate for the loss in interference. On the other hand, by Nactive-relays = pactive Nrelay . In order to keep the costs at a
the smart antenna scheme will naturally provide improved minimum, the number of base stations Nbase is selected so
spatial multiplexing possibilities with an increased number of that the ratio Nactive-relay /Nbase attains exactly the maximum
antennas, which leads to an improved system throughput. capacity, K (in terms of relays per site). By the simulations
Another important aspect is the fairness between the users. in the previous section for any selection of antennas in the
In Figures 4 and 5, the fairness is illustrated by plotting the base- and mobile-station we do a minimum dimensioning, i.e.
cumulative density (over users and scenarios) of the average K = Nactive-relay /Nbase . The total cost for base-stations, relays
data rate a particular user perceives in a particular scenario. and user equipments is then given by
Naturally, the round-robin based scheduling and the attempt Ctotal = Cbase (Nt )Nbase + Crelay (Nr )Nrelay + CUE NUE (1)
to provide the same average SINR for each user in the smart
antenna scheme provides higher rate for the users with worst where Cunit is the cost per unit and the dependence of the costs
channel conditions. Still, it does not provide exactly the same on the number of antennas (Nt and Nr ) has been emphasized.
rate to all users, since adaptive modulation is used and the As the criterion function for our further analysis we will be
scheduling and transmit beamforming is only based on the the cost per user equipment i.e. CUE = Ctotal /NUE . Combining
long-term channel information. Note that all the different the equations above yields
schemes contain tuning parameters that influence the trade-off pUE-active Crelay (Nr )
between fairness and system throughput, so the results may Cuser = Cbase (Nt )( + ) + CUE (2)
K Cbase NUE-per-relay
change with other choices of parameters. pUE-active
= Cbase (Nt )( + f ) + CUE (3)
K
V. C OST A NALYSIS
where “the fractional relay cost”, f , is given by
In this section, we try to incorporate the performance figures
Crelay (Nr )
from the previous section into a model for the cost per user f= . (4)
equipment. In order to get a numerical cost estimate, we use Cbase NUE-per-relay
a combination of cost figures from the literature and educated Equation (3) allows an easy interpretation for a fixed base
guesses in Section V-B. station cost. The first term is the cost for the base station part
- split equally over all user equipments, while the second term
A. General is the fractional mobile relay station cost (relative to the base
We assume a service area with a number user equipments to station) divided by the number of user equipments who share
be served by an infrastructure consisting of mobile relays and it. The equation becomes more favorable the more users that
base stations. The number of user equipments is given by NUE can share the same relay. It is possible to see the trade-off
and they are active with a probability pUE-active . The number between the relay and base station costs. By increasing the
of users sharing a mobile-relay to access the network is given number of mobile-station antennas the K factor will increase
by NUE-per-relay and thus the number of mobile relays, Nrelay , and thus the base station cost is reduced, on the other hand
is given by Nrelay = NUE /NUE-per-relay . We assume that not the cost for the mobile relays is increased, and vice versa.
Nr Nt Smart Opport.
B. Numerical Prediction 1 2 NA NA
2 2 1.65 3.6
To be able to provide a numerical comparison, the following 4 2 1.1 2.03
assumptions are used: 8 2 1.08 1.41
1 4 1.74 NA
• The relation of the cost of a multi-antenna relay and that
2 4 1.21 NA
of an IEEE802.11a modem is as in [14]. However it is 4 4 1.04 NA
modified with the matrix inversion rate assumed here i.e. 8 4 0.73 NA
once for each sub-carrier at every timeslot of 0.123ms. HSDPA 2.98
• The cost of a HSDPA modem is assumed to be two times TABLE II
the cost of an IEEE802.11a modem (in [14] it is assumed C OST PER USER EQUIPMENT [ K EUR] FOR THE S MART AND
to be five times as expensive). O PPORTUNISTIC SCHEMES .
• The cost of an IEEE802.11a modem is 25 Euro.
• The cost of the relay to user equipment links is assumed
equal to the cost of an IEEE802.11a modem.
• The BS cost is that given in [15] with the multi-antenna R EFERENCES
modeling of [16]. [1] K. Johansson, J. Markendahl, and P. Zetterberg, “Relaying access points
• The probability of a user equipment being active is and related business models for low cost mobile systems,” Austin
Mobility Roundtable, Mar. 2004.
pUE-active = 1%. [2] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
• The capacity, K, is the mean number of relays per site Communications. Cambridge, United Kingdom: Cambridge University
for which the user data-rate is above 2.0Mbit/s with 90% Press, 2003.
[3] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
probability and 11.0Mbit/s with 50% probability. This is Cambridge University Press, 2005.
a favorable selection for HSDPA. [4] P. Viswanath, D. N. C. Tse, and R. Laroia, “Opportunistic beamforming
• Number of UE per relay NUE-per-relay = 5. using dumb antennas,” IEEE Trans. IT, vol. 48, no. 6, pp. 1277–1294,
June 2002.
Based on these assumptions, the cost per user, Cuser , in kEUR [5] M. Sharif and B. Hassibi, “On the capacity of MIMO broadcast channels
with partial side information,” IEEE Trans. IT, vol. 51, no. 2, pp. 506–
is obtained as given in Table II below. Note that for a given 522, Feb. 2005.
number of base- and mobile relay antennas the cost of the [6] P. Svedman, L. J. Cimini Jr., M. Bengtsson, S. K. Wilson, and
smart and opportunistic scheme are the same, thus the system B.Ottersten, “Exploiting temporal channel correlation in opportunistic
SD-OFDMA,” in Proc. ICC, 2006, accepted.
providing the largest capacity gain will have the lowest cost. [7] M. Bengtsson, P. Svedman, X. Zhang, and P. Zetterberg, “System
This is because the analysis is not so fine-grained that the comparison of smart and dumb antennas,” in Proc. VTC 2005 Spring,
computational load in the base-station matters. On the other June 2005.
[8] M. Bengtsson, “Pragmatic multi-user spatial multiplexing with robust-
hand the computational load of matrix inversions is a major ness to channel estimation errors,” in Proc. IEEE ICASSP, vol. IV, Apr.
cost driver but this is the same in both systems - here is a big 2003, pp. 820–823.
potential for cost savings. In the table NA means not applicable [9] F. Shad, T. D. Todd, V. Kezys, and J. Litva, “Dynamic slot allocation
(DSA) in indoor SDMA/TDMA using a smart antenna basestation,”
- which happens when the required user data-rate was not met IEEE/ACM Trans. on Networking, vol. 9, no. 1, pp. 69–81, Feb. 2001.
at any load. As indicated, the lowest cost was obtained using [10] 3GPP TR 25.996 Spatial Channel Model for Multiple Input Multiple
the smart antenna scheme with four base-station and eight Output (MIMO) Simulations, 3GPP; Technical Specification Group
Radio Access Network, rel. 6.
mobile-station antennas. For this constellation the capacity [11] J. Salo, G. Del Galdo, J. Salmi, P. Kyösti, M. Milojevic, D. Laselva,
gain over HSDPA is 6.4 while the cost "loss" is 4.1. This and C. Schneider. (2005, Jan.) MATLAB implementation of the
means that part of the capacity gain is “eaten” by the cost 3GPP Spatial Channel Model (3GPP TR 25.996). [Online]. Available:
http://www.tkk.fi/Units/Radio/scm/
associated with the solution. The opportunistic scheme always [12] S. Kandukuri and S. Boyd, “Optimal power control in interference-
performs worse since it provides less capacity gain. This is limited fading wireless channels with outage-probability specifications,”
probably due to the low average number of relays per sector, IEEE Trans. Wireless Comm., vol. 1, no. 1, pp. 46–55, Jan. 2002.
[13] W. Anchun, X. Liang, Z. Shidong, X. Xibin, and Y. Yan, “Dynamic
which results in a low multiuser diversity. resource management in the fourth generation wireless systems,” in
Proc. Int.. Conf. Communication Technology, vol. 2. IEEE, 2003, pp.
VI. C ONCLUSIONS 1095–1098.
[14] P. Zetterberg, “A smart antenna concept for 4g and transceiver cost,” in
Proceedings of Nordic Radio Symposium, Aug. 2004.
This paper presents an attempt to compare a smart antenna [15] K. Johansson, A. Furuskär, P. Karlsson, and J. Zander, “Relation between
solution based on round-robin scheduling with an opportunis- cost structure and base station characteristics in cellular systems,”
tic beamforming solution using proportional fair scheduling. IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications, Sept. 2004.
The comparison considers not only system throughput and [16] J. Markku, J. Heikkil, J. Ylitalo, E. Tiirola, and A. Pollard, “Architec-
user fairness but also system cost, assuming that the mobiles tural design and cost impact,” IST I-METRA, Tech. Rep. D5.1, 2003,
are used as mobile relays. Making a fair comparison between available at http://www.ist-imetra.org/deliverables/NMP-WP5-D5.1-V1.
1.pdf.
smart and opportunistic multi-antenna schemes is a difficult
task, but we have highlighted some important aspects. For
the low-load low-speed scenario we have considered, a smart
antenna scheme would be preferable in terms of system cost.