WRN: improving system performance in 3G networks ... - IEEE Xplore
WRN: Improving System Performance in 3G Networks Through Fixed Multi-hop Relay Nodes Shamik Sengupta and Mainak Chatterjee
Samrat Ganguly and Rauf Izmailov
Electrical and Computer Engineering University of Central Florida Orlando, FL 32816-2450 Email: {shamik, mainak}@cpe.ucf.edu
Broadband and Mobile Communications NEC Laboratories America Princeton, NJ 08540 Email: {samrat, rauf}@nec-labs.com
Abstract— Non-uniform coverage and low system throughput due to location dependent fading in cellular networks is a major bottleneck in providing quality of service. Moreover, the growing number of stand-alone Wi-Fi hotspots is creating competition for the 3G cellular networks. In this paper, we propose an architecture which augments the existing 3G downlink technology with Wi-Fi like relay nodes leveraging benefits of both forms to improve the 3G system performance and provide uniform coverage throughout the cell. The relay nodes in the proposed WRN (Wi-Fi like Relay Network) architecture are equipped with dual radio interface to communicate with the base station and other relay nodes. We compare the performance of the existing CDMA/HDR-based 3G cellular network with the proposed WRN architecture in terms of throughput and blocking probability. We demonstrate how pre-engineered deployment of relay nodes can yield better performance than random deployment of relay nodes. Simulation experiments are conducted to corroborate our analytical findings. The system throughput is not only enhanced but the fairness among users is also maintained.
I. I NTRODUCTION Radio waves experience path loss which is inversely proportional to the distance traveled raised to some exponent as they propagate from a transmitter to a receiver. This becomes a serious issue in wireless cellular environment where mobile terminals are at varying distances from the base station, as a consequence of which a nearby terminal will have more power transfer than a one which is at a further distance. Distance dependent power reception coupled with the effects due to building and topological profile, results in non-uniform coverage within a cell. This kind of poor coverage is a major problem because it threatens QoS provisioning. To cope with the growing needs of data services, 3GPP has defined high data rate (HDR) [3] and high speed data packet access (HSDPA) [1]. For such services, the same notion of power sharing used for voice services no longer remains valid because packet data systems are aimed at maximizing the throughput among other parameters for QoS. Furthermore, since HDR and HSDPA do not have power control, users may experience reduced downlink rate based on their location in the cell. This problem cannot be solved completely at the base station through scheduling or modulation techniques. If the scheduler tries to provide fair downlink rate to users independent of their locations, the overall throughput will be compromised; if the scheduler tries to maximize overall
IEEE Communications Society / WCNC 2005
throughput, fairness will be compromised. The real challenge is not to find a better tradeoff between fairness and maximizing throughput, but instead find a solution where fair service for users is accomplished with efficient network utilization. The concept of using mobile terminals to relay traffic for other users have been considered in [2], [8]. This kind of model lacks the sense of economic incentives. Use of dedicated relay nodes without involving the mobile users terminals have been considered in [9], but primarily for voice traffic. Furthermore, most of the schemes [7], [5], [2], [8] use the same channel for both access control and traffic relaying; as a result, they suffer from interference. In contrast to the previous work, the goal of this paper is to provide uniform coverage within a single cell. The proposed architecture uses the Wi-Fi spectrum for relaying, which causes no interference for the 3G channels. We propose an architecture of augmenting the 3GPP mechanism based on CDMA/HDR with a multi-hop relay network based on the WiFi standard to create a Wi-Fi like Relay Network, where the relay nodes bring the data from distant base station closer to the user and thus improve the cellular coverage using radio resources that do not belong to the 3G network. In WRN, some relay nodes (RN) are equipped with dual interface (3G and Wi-Fi) while others have only Wi-Fi interface. We assume that each mobile node (MN) has dual interface in order to use the service provided by the WRN. Though we consider HDR, the approach is generic enough to be applied to any 3G standard. This flexible architecture provides higher overall downlink data service to users independent of his location with low cost opportunistic deployment of RNs operating on Wi-Fi-based infrastructure. For service providers, this architecture has the flexibility to define the usage of the RNs as a new service and charging the user for delivering this service. The rest of the paper is organized as follows. In section II, we explain the existing time slotted CDMA/HDR system and analyze the system performance based on different scheduling policy. In section III, we introduce WRN and provide the implementation guidelines. Performance analysis is done from throughput and blocking probability point of view for the proposed architecture in section IV. We present the simulation results in section V. Conclusions are drawn in section VI.
1708
0-7803-8966-2/05/$20.00 © 2005 IEEE
It can be noted that the throughput is not dependent on the number of users in the system. More users in the cell would simply mean that each user would get less to account for the capacity constraint of the base station. Thus, the throughput of an individual user for various distances of the user from the base station in a system of , where, Tj is N active users can be given by, Tj = kCi i=1
System throughput with round robin scheduling System throughput with Opportunistic scheduling
2500
2000
1500
1000
500
0
Fig. 1.
0
1000
500
1500 2000 Radius of the cell (meter)
2500
3000
System Throughput for round robin and opportunistic scheduling 8 Individual user Throughput with round robin scheduling Individual user Throughput with Opportunistic scheduling
7
6
5
4
3
2
1
0 1000
Fig. 2.
1200
1400
1600
2200 1800 2000 Distance of a User (meter)
2400
2600
2800
3000
Per-user Throughput for round robin and opportunistic scheduling
Though opportunistic scheduling is successful and widely used, it fails to address the issue of fairness as evident from figure 2. We observe that the per-user throughput is worse when the number of users in a cell increases. So, we introduce the concept of WRN to improve this scenario and to obtain highest possible system throughput and per-user throughput at the same time. III. P ROPOSED WRN A RCHITECTURE
ρAi
the individual throughput of jth user, when this jth user is in the ith ring and is receiving data at Ci Kbit/sec. Opportunistic scheduling: In round robin scheduling, as users were given same number of slots regardless of their received data rates and locations, system throughput was reduced with more number of users being at the edge of the cell. In opportunistic scheduling policies, users with higher
IEEE Communications Society / WCNC 2005
3000
System Throughput (Kbit/s)
Let us briefly discuss the limitations in multi-rate CDMA systems in this section and demonstrate the effect of fading and discrete data rates on the system throughput. In the existing time-slotted CDMA/HDR system where the base station transmits at full power to only one user in each slot, the cell is divided into 11 concentric rings each receiving a particular data rate due to coding, distance and fading constraints of HDR. Since, this research is not constraint by HDR technology, we consider a system with k different rates. Thus, for the k discrete rates, the cell would have k concentric rings. The data rate for the innermost ring is the maximum and it decreases for outward rings. All users belonging to a particular ring is assumed to get the same data rate. To analyze this system, we consider a cell of radius R with users uniform randomly distributed over the entire cell. Let ρ be the density of active users and A1 , A2 , · · · , Ai , · · · , Ak be the areas of the 1, 2, · · · , i, · · · , k rings respectively. The number of users in ith ring is ρAi . The expected number of ktotal active users (N ) in the cell can then be given by i=1 ρAi . We assume data rate perceived by a user in the ith ring as Ci Kbit/sec (i = 1, · · · , k). Then we analyze this existing time-slotted CDMA/HDR system from two scheduling point of view: simple round robin scheduling, where the fairness criteria is maintained but the system throughput is compromised and the opportunistic scheduling where the system throughput is better than the idealistic round robin policy but the users near the edge of the cell suffer the most. Round-robin scheduling: We consider a cell with N active users as mentioned before. Then in an idealistic scenario of round robin scheduling policy, where each user gets an equal number of slots, after N slots, every active user would have obtained exactly one slot. The total amount kof data received by the users in these N slots is given by i=1 ρAi Ci τ where, duration of each slot is given by τ seconds. Therefore, the throughput of the system (T ) is obtained by dividing the total data transferred by the time duration of N slots. Thus, k k i=1 ρAi Ci τ i=1 ρAi Ci = (1) T = k k τ × i=1 ρAi i=1 ρAi
perceived data rates are given priorities than the users with poor perceived data rates to enhance the system throughput. But this in turn reduces the individual user throughput for the users near the edge of the cell, who are receiving poor data rate due to fading. We analyzed one such opportunistic scheduling policy, in which users with higher data rates were given more number of slots. We consider the perceived rates of HDR as shown in Table I [4] and N = 200. In figures 1 and 2, we compare the system throughput and individual user throughput for round robin and opportunistic scheduling. We observe a clear trade-off in these results- better system throughput is achieved with opportunistic scheduling, however individual user throughput reduces drastically with increase of distance.
Individual user Throughput (Kbit/s)
II. L IMITATIONS IN E XISTING CDMA/HDR S YSTEM
The three main entities in WRN are 3G base station (BS), relay nodes (RN) and mobile nodes (MN) such as multimedia phones and notebooks. The 3G BS with fixed location covers a cell and provides downlink data channel access to the MNs. The currently used technologies (HDR and HSDPA) for high downlink data access do not have power control capabilities, thus the achievable rates vary depending on the user location.
1709
0-7803-8966-2/05/$20.00 © 2005 IEEE
3G BS
3G Wireless Transmission Wi−fi like Wireless Transmission Relay Node 3G cellular Network Converted to Wi−fi like Relay Network (WRN)
Fig. 3.
Mobile Node
WRN architecture
To confront this situation, WRN is composed of RNs which can function as relays to forward the downlink data from the 3G BS. These RNs can be deployed in ad-hoc or preplanned manner and can also be shifted or moved based on the spatial change in traffic/user concentration. We also assume that these RNs are equipped with a Wi-Fi like interface in adhoc mode and they have a definite transmission range. We call this area within the transmission range of a RN as relay cell. A MN must be within the transmission range of at least one RN to receive data from RN. However there are certain RNs which have dual interface (3G and Wi-Fi like), and they are capable of receiving downlink data directly from the 3G BS and forward it to either MNs or other RNs. The other type of RNs have Wi-Fi like interface only and they are capable of receiving downlink data from other RNs only and forward it to either MNs or RNs. Each mobile node, which wants to take advantage of the proposed architecture is equipped with dual interface. We categorize the downlink data forwarding path from 3G BS to the MN as direct and indirect. In direct forwarding, the MN receives data directly from BS using the 3G interface; on the other hand, in the indirect forwarding, the data forwarding path is through the RNs and data at MN is received through the Wi-Fi interface. Note that MNs are disabled for relaying and therefore cannot relay data to other MNs. This avoids extra battery power consumption by the mobile terminals. One of the main goals of this proposed architecture is to provide flexibility in deployment of the WRN. Attaining the flexibility necessitates that the WRN is self-organizing. When a set of RNs is deployed covering certain geographical area, the RNs should be able to self configure and organize to form the WRN and become capable of relaying the downlink data from 3G BS. Therefore, the routing protocol should include a discovery process to allow for the self-organizing nature. In the current architecture and application, one needs to observe the level of mobility of each nodes participating in forwarding and receiving downlink packets. Obviously, the BS is always fixed. The RNs are quasi-stationary implying that they can be shifted, added, or deleted. Nevertheless, the time interval required for topology update in the WRN can safely be assumed to be longer than most mobile client session duration. For routing mechanisms, the knowledge required by the BS is the identity of the RN that acts as a proxy client on behalf of a given MN. Based on this knowledge, the radio network controller (RNC) requires the only functionality that
IEEE Communications Society / WCNC 2005
enables shifting the traffic from direct to indirect path and vice versa to a given MN based on need. For each registered MN, RNC maintains a proxy client which is either flushed or updated based on report from the WRN. Through uplink channel, the MN notifies the RNC if it wants the downlink traffic forwarded through direct or indirect path. When the indirect path is chosen, the IP packets are encapsulated with destination address of the proxy clients (RN) id at RNC. Upon receiving the IP packets through the 3G interface at RN, the IP packets are decapsulated and forwarded to WRN based on the routing protocol. Moreover, since the bandwidth of the RNs is orders of magnitude greater than that of the BS, it can be argued that there would be no queuing delay at the RNs. IV. A NALYTICAL M ODELING The performance of the WRNs depends heavily on the placement of the relay nodes. We cannot place the relay nodes anywhere in the cell because that might introduce huge overlaps and might render certain relay nodes redundant. But as the transmission range of relay nodes is not much, so overlapping must be there for multi-hop transmissions to connect the relay nodes. So, to get the maximum out of each relay node, it is necessary that they be deployed in a pre-engineered manner. Pre-engineered Deployment: It is well known from planar geometry that to cover a 2-dimensional region with equal sized circles, the best possible packing can be obtained by surrounding each circle by six circles as shown in figure 4. But to have connection between relay nodes, we need overlap also to some extent between relay cells. We therefore consider a situation where the location of the relay nodes are pre-computed and the best deployment strategy is chosen. The deployments shown in figure 4 are just two examples of such pre-engineered approach with certain number of relay nodes. The first deployment tries to cover the entire cell, while the second one tries to cover a densely populated region of MNs near the edge of the cell. Random Deployment: Since the pre-engineered deployment will not be possible everywhere due to accessibility problem in RN sites, random deployment is an alternative solution to this problem. In the simulation results section we analyze the WRN performance based on random deployment and compare both the deployment strategies. A. Throughput We define system throughput as the total data (in Kbits) received by users per unit time considering the effects due to fading and shadowing. We consider a cell with N active users as before. Let there be L relay nodes each of which is forming relay cell of radius r. If L is sufficiently large such that all the relay nodes entirely cover the cell, then the expected number of users under one relay node is N/L. However, if L is small, all the users might not be covered by the relay nodes as the coverage area of the relay nodes is usually small. But users covered by a relay node get the highest data rate possible.
1710
0-7803-8966-2/05/$20.00 © 2005 IEEE
relay nodes, system throughput also increases continuously, till a certain value of L is reached, where the maximum value of throughput is obtained. In an ideal situation, this value of L is the saturation point after which the throughput does not increase since all the users are accommodated by the relay nodes. Fig. 4.
Examples of pre-engineered placement of relay nodes
Let us define ar as the area of the relay cell. Therefore, the expected number of users that fall within the L relay cells are ρar L. These ρar L users then have the choice of choosing either the direct 3G path from the BS or indirect path through RNs depending on the signal strength. Let us assume that ρar L < N , i.e., not all users belong to relay cells. The remaining (N − ρar L) active users then must receive their signals from the BS through the direct path during their allocated time slot. If we assume idealistic round robin scheduling policy to maintain the fairness criteria, then after N slots each user would receive exactly one slot. As a result, throughput for the system with L relay nodes can be given by, N −ρa L ρar LCτ + j=1 r Cj τ (2) TRS = N where τ is the duration of each slot, C is the highest data rate at which the relay nodes are transmitting and Cj is the 3G data rates received by the MNs which are not covered by the RNs. Now as we assume that the data transmission between relay nodes and users take the help of Wi-Fi like technology and r is much lesser than R, then there is very little probability of any loss of signal strength. As a result, without any loss of generality, we assume that users under the direct influence of relay nodes, get data at the highest possible rate C. For analysis, we consider a single-cell system with N as 1000. The radius of the cell and the relay cells are considered 2500 and 500 meters respectively. We considered received data rates and corresponding distances from Table I as given in [4]. We increased the number of relay nodes gradually in equation (2) to evaluate the system throughput. 3000 throughput with relay stations − pre−engineered approach Throughput without relay stations − round robin scheduling Throughput without relay stations − opportunistic scheduling
System Throughput (Kbit/s)
2500
2000
1500
1000
500
0
Fig. 5.
0
10
20
30
60 50 40 no. of relay stations
70
80
90
B. Blocking Probability We define blocking probability as the probability with which a user’s request is rejected i.e., the request is denied the permission to even enter the waiting queue. Let us first consider a system without relay nodes where each user is guaranteed a minimum data rate regardless of his location. If the server has enough bandwidth available to serve all the existing users along with the new user providing at least their guaranteed data rates to all of them, then only a new user is admitted into the server. Otherwise, the new user is requested to stay in the queue, for some time, after which, upon availability of the bandwidth the new user is admitted. We assume a queue of size M . We denote the data rate perceived by a user in the ith ring in his/her own slot as Ci Kbit/sec (i = 1, · · · , k) and the guaranteed minimum data rate provided to all the users as Cmin Kbit/sec for the entire time, where Cmin S We assume every user intends to download file of size F . Note: µS is not a constant but rather depends on the perceived data rates of S users being served in the server in time slotted mechanism. Then, the equilibrium probabilities P (j) are related to each other through the following state balance equations: µj P (j) = λP (j − 1) if 1 ≤ j ≤ S (5) µS P (j) = λP (j − 1) if S < j ≤ S + M
= P (S + M ) =
Using the above equations recursively along with the normalS+M ization condition we have, j=0 P (j) = 1. Then the steady state probabilities can be given by, j P (0) if 1 ≤ j ≤ S jλ µi i=1 P (j) = (6) j λ P (0) if S < j ≤ S + M j j−S (µS )
(
i=1
Ring k
Data rate ( Kbit/sec )
Radius (m.) (α = 4)
11 10 9 8 7 6 5 4 3 2 1
38.4 76.8 102.6 153.6 204.8 307.2 614.4 921.6 1228.8 1843.2 2457.6
2820 2370 2210 2000 1860 1680 1410 1280 1190 1070 1000
µi )
TABLE I R ATES AND R ING R ADIUS
where,
In figure 7, we plot the blocking probability against arrival rate graphs for with and without RNs as derived from (7) equations (8) and (10). As evident from the figure, blocking j=1 probability decreases with introduction of RNs. Of course, the Based on the above analysis and probability definitions, we cost of deploying the RNs is not considered here. can define P (S + M ) as the Blocking Probability (BP ). Then blocking probability of this time-slotted CDMA/HDR model can be given as,
S λj P (0) = 1 + j
S+M
λj + j−S ( j i=1 µi i=1 µi ) j=S+1 (µS )
−1
Without Relay Station With Relay Station
0.5
= P (S + M ) =
λS+M P (0) S (µS )M ( i=1 µi )
0.4
(8)
Let us now analyze the system with the relay nodes. As the relay nodes are just overlaid on the existing system model without RNs, we keep all the aspect of the previous model fixed but the service rate. As the RNs transmit at a short range with Wi-Fi like mechanism, loss due to fading and shadowing can be assumed almost negligible and data rates perceived by the users are the maximum possible for the users under the direct influence of the RNs. Then the service rate in this model can be given by, j Cmax i=1 if j ≤ S jF (9) µj = µS if j > S where, Cmax is the maximum possible data rate perceived by the users under the direct influence of the RNs and moreover,
IEEE Communications Society / WCNC 2005
Blocking Probability
BP
0.3
0.2
0.1
0 0
Fig. 7.
0.05
0.1
0.15
0.2 0.25 Arrival Rate
0.3
0.35
0.4
0.45
Blocking Probability with and without relay nodes V. S IMULATION R ESULTS
To verify our proposed model we performed simulation experiments on a UNIX based platform. We considered that the users are scattered randomly over the cell. For the relay nodes, we considered both situations - randomly placed and pre-engineered. Random placement would incur overlaps of relay cells whereas the pre-engineered would not till a certain number of relay nodes.
1712
0-7803-8966-2/05/$20.00 © 2005 IEEE
A. Throughput vs. number of relay nodes We considered N (number of active users) as 1000, radius of the cell (R) as 2500m and radius of the relay cell (r) as 500m. The throughputs are shown in figure 8. For the preengineered case, the throughput is almost the same as was obtained analytically. With increase of relay nodes, more users come under the direct influence of relay nodes. When all the users are covered by the relay nodes, the system throughput reaches its maximum value.
relay nodes does not vary much. This is because all the users are close to the base station and receive very high rate. But when the cell radius increases, more and more number of users get lesser signal strength and hence lesser data rate, resulting in decreased system throughput. With the presence of relay nodes, the far off users are virtually brought near the base station allowing them to receive higher data rate. The figure also shows that for covering a larger cell, more relay nodes would be required if the same throughput is to be expected. 3000
3000
throughput with 40 relay stations throughput with 25 relay stations throughput without relay
2500
2500
2000
2000
Throughput (Kbit/s)
System Throughput (Kbit/s)
throughput with relay stations − pre−engineered approach throughput with relay stations − random approach Throughput without relay stations − opportunistic scheduling policy Throughput without relay stations − round robin scheduling policy
1500
1000
1000
500
500
0
0 0
10
Fig. 8.
20
30
60 50 40 no. of relay stations
80
70
90
100
22 throughput with 40 relay stations throughput with 25 relay stations throughput without relay
18
1000
2500 2000 1500 Radius of the cell (meter)
3000
3500
4000
Effect of radius ratio
In this paper, we proposed augmentation of 3G cellular networks with Wi-Fi like relay nodes to provide users with high data rates regardless of their location with respect to the base station. We analytically showed how uniform coverage within a cell can be obtained by the introduction of relay nodes. We also showed how pre-engineered positioning of the RNs can provide better performance compared to random placement. We performed simulation experiments with data rates of HDR to evaluate the throughput and blocking probability of the system as well as per-user throughput for both the deployment strategies. It is shown that the use of relay stations always enhances the performance, relatively better if pre-engineered.
16
R EFERENCES
14
12
10 8 6 4 2
0
500
VI. C ONCLUSIONS
Here, we study the per-user throughput for both with and without relay nodes. We simulated with the same values for R and r as mentioned earlier but considered two different number of relay nodes - 25 and 40. The results obtained are shown in figure 9. As expected, with the increase in number of active users, the per-user throughput decreases because of resource sharing, but the rate of decrease is lesser for the case with relay nodes than without relay nodes. With increased number of relay nodes, this decreasing rate can be further reduced. 20
0
Fig. 10.
Throughput: random and pre-engineered
B. Per-user Throughput vs. number of users
Throughput per user (Kbit/s)
1500
0
200
Fig. 9.
400
600
800
1200 1000 no. of users
1400
1600
1800
2000
per-user Throughput: random deployment
C. Throughput vs. radius of cell So far, the radii of the cell and the relay nodes were considered constant. Though the absolute value is of less importance, what is decisive is the ratio of the two radii. The results obtained so far assumed a ratio of 1 : 5 (500m : 2500m). We keep the radius of the relay node constant (500m) and vary the radius of the cell to 4000m so as to get a varying radius ratio. We considered the same number of relay nodes25 and 40. The throughputs with and without relay nodes with varying cell radius are shown in figure 10. We observe that when the radius (R) is small, throughput with and without
IEEE Communications Society / WCNC 2005
1713
[1] 3GPP TR25.848 V0.5.0, “Physical Layer Aspects of UTRA High Speed Downlink Packet Access(HSDPA),” ANNEX B, TSGR1#18(01)186, Jan 18,2001. [2] G. Aggelou and R. Tafazolli, “On the relaying capacity of nextgeneration GSM cellular networks,” IEEE Personal Communications Magazine, 8(1):40-47, Feb. 2001. [3] P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushyana, S. Viterbi, “CDMA/HDR: a bandwidth efficient high speed wireless data service for nomadic users”, IEEE Communications Magazine, Volume: 38 Issue: 7, July 2000, pp. 70-77. [4] T. Bonald and A. Proutiere, “Wireless Downlink Data Channels: User Performance and Cell Dimensioning”, Proceedings of ACM Mobicom, 2003, pp. 339-352. [5] H. Hsieh and R. Sivakumar, “On Using the Ad-hoc Network Model in Wireless Packet Data Networks,”In Proceedings of ACM MobiHoc,2002. [6] L. Kleinrock and J. Silvester, “Optimum transmission radii for packet radio networks or why six is a magic number,” Proc. IEEE National Telecommunications Conference, 1978, pp. 4.3.1-4.3.5. [7] Y. Lin and Y. Hsu, “Multihop Cellular: A New Architecture for Wireless Communications,” In Proceedings of IEEE INFOCOM, volume 3, 2000. [8] H. Luo, R. Ramjeey, P. Sinha, L. Li, S. Lu, “UCAN: A Unified Cellular and AdHoc Network Architecture,” In proceedings of Mobicom, 2003. [9] H. Wu, C. Qiao, S. De and O. Tonguz, “iCAR: Integrated cellular and ad hoc relaying systems,” IEEE Journal on Selected Areas in Communications, Volume: 19, Issue: 10, pp. 2105-2115.
0-7803-8966-2/05/$20.00 © 2005 IEEE
Samrat Ganguly and Rauf Izmailov
Electrical and Computer Engineering University of Central Florida Orlando, FL 32816-2450 Email: {shamik, mainak}@cpe.ucf.edu
Broadband and Mobile Communications NEC Laboratories America Princeton, NJ 08540 Email: {samrat, rauf}@nec-labs.com
Abstract— Non-uniform coverage and low system throughput due to location dependent fading in cellular networks is a major bottleneck in providing quality of service. Moreover, the growing number of stand-alone Wi-Fi hotspots is creating competition for the 3G cellular networks. In this paper, we propose an architecture which augments the existing 3G downlink technology with Wi-Fi like relay nodes leveraging benefits of both forms to improve the 3G system performance and provide uniform coverage throughout the cell. The relay nodes in the proposed WRN (Wi-Fi like Relay Network) architecture are equipped with dual radio interface to communicate with the base station and other relay nodes. We compare the performance of the existing CDMA/HDR-based 3G cellular network with the proposed WRN architecture in terms of throughput and blocking probability. We demonstrate how pre-engineered deployment of relay nodes can yield better performance than random deployment of relay nodes. Simulation experiments are conducted to corroborate our analytical findings. The system throughput is not only enhanced but the fairness among users is also maintained.
I. I NTRODUCTION Radio waves experience path loss which is inversely proportional to the distance traveled raised to some exponent as they propagate from a transmitter to a receiver. This becomes a serious issue in wireless cellular environment where mobile terminals are at varying distances from the base station, as a consequence of which a nearby terminal will have more power transfer than a one which is at a further distance. Distance dependent power reception coupled with the effects due to building and topological profile, results in non-uniform coverage within a cell. This kind of poor coverage is a major problem because it threatens QoS provisioning. To cope with the growing needs of data services, 3GPP has defined high data rate (HDR) [3] and high speed data packet access (HSDPA) [1]. For such services, the same notion of power sharing used for voice services no longer remains valid because packet data systems are aimed at maximizing the throughput among other parameters for QoS. Furthermore, since HDR and HSDPA do not have power control, users may experience reduced downlink rate based on their location in the cell. This problem cannot be solved completely at the base station through scheduling or modulation techniques. If the scheduler tries to provide fair downlink rate to users independent of their locations, the overall throughput will be compromised; if the scheduler tries to maximize overall
IEEE Communications Society / WCNC 2005
throughput, fairness will be compromised. The real challenge is not to find a better tradeoff between fairness and maximizing throughput, but instead find a solution where fair service for users is accomplished with efficient network utilization. The concept of using mobile terminals to relay traffic for other users have been considered in [2], [8]. This kind of model lacks the sense of economic incentives. Use of dedicated relay nodes without involving the mobile users terminals have been considered in [9], but primarily for voice traffic. Furthermore, most of the schemes [7], [5], [2], [8] use the same channel for both access control and traffic relaying; as a result, they suffer from interference. In contrast to the previous work, the goal of this paper is to provide uniform coverage within a single cell. The proposed architecture uses the Wi-Fi spectrum for relaying, which causes no interference for the 3G channels. We propose an architecture of augmenting the 3GPP mechanism based on CDMA/HDR with a multi-hop relay network based on the WiFi standard to create a Wi-Fi like Relay Network, where the relay nodes bring the data from distant base station closer to the user and thus improve the cellular coverage using radio resources that do not belong to the 3G network. In WRN, some relay nodes (RN) are equipped with dual interface (3G and Wi-Fi) while others have only Wi-Fi interface. We assume that each mobile node (MN) has dual interface in order to use the service provided by the WRN. Though we consider HDR, the approach is generic enough to be applied to any 3G standard. This flexible architecture provides higher overall downlink data service to users independent of his location with low cost opportunistic deployment of RNs operating on Wi-Fi-based infrastructure. For service providers, this architecture has the flexibility to define the usage of the RNs as a new service and charging the user for delivering this service. The rest of the paper is organized as follows. In section II, we explain the existing time slotted CDMA/HDR system and analyze the system performance based on different scheduling policy. In section III, we introduce WRN and provide the implementation guidelines. Performance analysis is done from throughput and blocking probability point of view for the proposed architecture in section IV. We present the simulation results in section V. Conclusions are drawn in section VI.
1708
0-7803-8966-2/05/$20.00 © 2005 IEEE
It can be noted that the throughput is not dependent on the number of users in the system. More users in the cell would simply mean that each user would get less to account for the capacity constraint of the base station. Thus, the throughput of an individual user for various distances of the user from the base station in a system of , where, Tj is N active users can be given by, Tj = kCi i=1
System throughput with round robin scheduling System throughput with Opportunistic scheduling
2500
2000
1500
1000
500
0
Fig. 1.
0
1000
500
1500 2000 Radius of the cell (meter)
2500
3000
System Throughput for round robin and opportunistic scheduling 8 Individual user Throughput with round robin scheduling Individual user Throughput with Opportunistic scheduling
7
6
5
4
3
2
1
0 1000
Fig. 2.
1200
1400
1600
2200 1800 2000 Distance of a User (meter)
2400
2600
2800
3000
Per-user Throughput for round robin and opportunistic scheduling
Though opportunistic scheduling is successful and widely used, it fails to address the issue of fairness as evident from figure 2. We observe that the per-user throughput is worse when the number of users in a cell increases. So, we introduce the concept of WRN to improve this scenario and to obtain highest possible system throughput and per-user throughput at the same time. III. P ROPOSED WRN A RCHITECTURE
ρAi
the individual throughput of jth user, when this jth user is in the ith ring and is receiving data at Ci Kbit/sec. Opportunistic scheduling: In round robin scheduling, as users were given same number of slots regardless of their received data rates and locations, system throughput was reduced with more number of users being at the edge of the cell. In opportunistic scheduling policies, users with higher
IEEE Communications Society / WCNC 2005
3000
System Throughput (Kbit/s)
Let us briefly discuss the limitations in multi-rate CDMA systems in this section and demonstrate the effect of fading and discrete data rates on the system throughput. In the existing time-slotted CDMA/HDR system where the base station transmits at full power to only one user in each slot, the cell is divided into 11 concentric rings each receiving a particular data rate due to coding, distance and fading constraints of HDR. Since, this research is not constraint by HDR technology, we consider a system with k different rates. Thus, for the k discrete rates, the cell would have k concentric rings. The data rate for the innermost ring is the maximum and it decreases for outward rings. All users belonging to a particular ring is assumed to get the same data rate. To analyze this system, we consider a cell of radius R with users uniform randomly distributed over the entire cell. Let ρ be the density of active users and A1 , A2 , · · · , Ai , · · · , Ak be the areas of the 1, 2, · · · , i, · · · , k rings respectively. The number of users in ith ring is ρAi . The expected number of ktotal active users (N ) in the cell can then be given by i=1 ρAi . We assume data rate perceived by a user in the ith ring as Ci Kbit/sec (i = 1, · · · , k). Then we analyze this existing time-slotted CDMA/HDR system from two scheduling point of view: simple round robin scheduling, where the fairness criteria is maintained but the system throughput is compromised and the opportunistic scheduling where the system throughput is better than the idealistic round robin policy but the users near the edge of the cell suffer the most. Round-robin scheduling: We consider a cell with N active users as mentioned before. Then in an idealistic scenario of round robin scheduling policy, where each user gets an equal number of slots, after N slots, every active user would have obtained exactly one slot. The total amount kof data received by the users in these N slots is given by i=1 ρAi Ci τ where, duration of each slot is given by τ seconds. Therefore, the throughput of the system (T ) is obtained by dividing the total data transferred by the time duration of N slots. Thus, k k i=1 ρAi Ci τ i=1 ρAi Ci = (1) T = k k τ × i=1 ρAi i=1 ρAi
perceived data rates are given priorities than the users with poor perceived data rates to enhance the system throughput. But this in turn reduces the individual user throughput for the users near the edge of the cell, who are receiving poor data rate due to fading. We analyzed one such opportunistic scheduling policy, in which users with higher data rates were given more number of slots. We consider the perceived rates of HDR as shown in Table I [4] and N = 200. In figures 1 and 2, we compare the system throughput and individual user throughput for round robin and opportunistic scheduling. We observe a clear trade-off in these results- better system throughput is achieved with opportunistic scheduling, however individual user throughput reduces drastically with increase of distance.
Individual user Throughput (Kbit/s)
II. L IMITATIONS IN E XISTING CDMA/HDR S YSTEM
The three main entities in WRN are 3G base station (BS), relay nodes (RN) and mobile nodes (MN) such as multimedia phones and notebooks. The 3G BS with fixed location covers a cell and provides downlink data channel access to the MNs. The currently used technologies (HDR and HSDPA) for high downlink data access do not have power control capabilities, thus the achievable rates vary depending on the user location.
1709
0-7803-8966-2/05/$20.00 © 2005 IEEE
3G BS
3G Wireless Transmission Wi−fi like Wireless Transmission Relay Node 3G cellular Network Converted to Wi−fi like Relay Network (WRN)
Fig. 3.
Mobile Node
WRN architecture
To confront this situation, WRN is composed of RNs which can function as relays to forward the downlink data from the 3G BS. These RNs can be deployed in ad-hoc or preplanned manner and can also be shifted or moved based on the spatial change in traffic/user concentration. We also assume that these RNs are equipped with a Wi-Fi like interface in adhoc mode and they have a definite transmission range. We call this area within the transmission range of a RN as relay cell. A MN must be within the transmission range of at least one RN to receive data from RN. However there are certain RNs which have dual interface (3G and Wi-Fi like), and they are capable of receiving downlink data directly from the 3G BS and forward it to either MNs or other RNs. The other type of RNs have Wi-Fi like interface only and they are capable of receiving downlink data from other RNs only and forward it to either MNs or RNs. Each mobile node, which wants to take advantage of the proposed architecture is equipped with dual interface. We categorize the downlink data forwarding path from 3G BS to the MN as direct and indirect. In direct forwarding, the MN receives data directly from BS using the 3G interface; on the other hand, in the indirect forwarding, the data forwarding path is through the RNs and data at MN is received through the Wi-Fi interface. Note that MNs are disabled for relaying and therefore cannot relay data to other MNs. This avoids extra battery power consumption by the mobile terminals. One of the main goals of this proposed architecture is to provide flexibility in deployment of the WRN. Attaining the flexibility necessitates that the WRN is self-organizing. When a set of RNs is deployed covering certain geographical area, the RNs should be able to self configure and organize to form the WRN and become capable of relaying the downlink data from 3G BS. Therefore, the routing protocol should include a discovery process to allow for the self-organizing nature. In the current architecture and application, one needs to observe the level of mobility of each nodes participating in forwarding and receiving downlink packets. Obviously, the BS is always fixed. The RNs are quasi-stationary implying that they can be shifted, added, or deleted. Nevertheless, the time interval required for topology update in the WRN can safely be assumed to be longer than most mobile client session duration. For routing mechanisms, the knowledge required by the BS is the identity of the RN that acts as a proxy client on behalf of a given MN. Based on this knowledge, the radio network controller (RNC) requires the only functionality that
IEEE Communications Society / WCNC 2005
enables shifting the traffic from direct to indirect path and vice versa to a given MN based on need. For each registered MN, RNC maintains a proxy client which is either flushed or updated based on report from the WRN. Through uplink channel, the MN notifies the RNC if it wants the downlink traffic forwarded through direct or indirect path. When the indirect path is chosen, the IP packets are encapsulated with destination address of the proxy clients (RN) id at RNC. Upon receiving the IP packets through the 3G interface at RN, the IP packets are decapsulated and forwarded to WRN based on the routing protocol. Moreover, since the bandwidth of the RNs is orders of magnitude greater than that of the BS, it can be argued that there would be no queuing delay at the RNs. IV. A NALYTICAL M ODELING The performance of the WRNs depends heavily on the placement of the relay nodes. We cannot place the relay nodes anywhere in the cell because that might introduce huge overlaps and might render certain relay nodes redundant. But as the transmission range of relay nodes is not much, so overlapping must be there for multi-hop transmissions to connect the relay nodes. So, to get the maximum out of each relay node, it is necessary that they be deployed in a pre-engineered manner. Pre-engineered Deployment: It is well known from planar geometry that to cover a 2-dimensional region with equal sized circles, the best possible packing can be obtained by surrounding each circle by six circles as shown in figure 4. But to have connection between relay nodes, we need overlap also to some extent between relay cells. We therefore consider a situation where the location of the relay nodes are pre-computed and the best deployment strategy is chosen. The deployments shown in figure 4 are just two examples of such pre-engineered approach with certain number of relay nodes. The first deployment tries to cover the entire cell, while the second one tries to cover a densely populated region of MNs near the edge of the cell. Random Deployment: Since the pre-engineered deployment will not be possible everywhere due to accessibility problem in RN sites, random deployment is an alternative solution to this problem. In the simulation results section we analyze the WRN performance based on random deployment and compare both the deployment strategies. A. Throughput We define system throughput as the total data (in Kbits) received by users per unit time considering the effects due to fading and shadowing. We consider a cell with N active users as before. Let there be L relay nodes each of which is forming relay cell of radius r. If L is sufficiently large such that all the relay nodes entirely cover the cell, then the expected number of users under one relay node is N/L. However, if L is small, all the users might not be covered by the relay nodes as the coverage area of the relay nodes is usually small. But users covered by a relay node get the highest data rate possible.
1710
0-7803-8966-2/05/$20.00 © 2005 IEEE
relay nodes, system throughput also increases continuously, till a certain value of L is reached, where the maximum value of throughput is obtained. In an ideal situation, this value of L is the saturation point after which the throughput does not increase since all the users are accommodated by the relay nodes. Fig. 4.
Examples of pre-engineered placement of relay nodes
Let us define ar as the area of the relay cell. Therefore, the expected number of users that fall within the L relay cells are ρar L. These ρar L users then have the choice of choosing either the direct 3G path from the BS or indirect path through RNs depending on the signal strength. Let us assume that ρar L < N , i.e., not all users belong to relay cells. The remaining (N − ρar L) active users then must receive their signals from the BS through the direct path during their allocated time slot. If we assume idealistic round robin scheduling policy to maintain the fairness criteria, then after N slots each user would receive exactly one slot. As a result, throughput for the system with L relay nodes can be given by, N −ρa L ρar LCτ + j=1 r Cj τ (2) TRS = N where τ is the duration of each slot, C is the highest data rate at which the relay nodes are transmitting and Cj is the 3G data rates received by the MNs which are not covered by the RNs. Now as we assume that the data transmission between relay nodes and users take the help of Wi-Fi like technology and r is much lesser than R, then there is very little probability of any loss of signal strength. As a result, without any loss of generality, we assume that users under the direct influence of relay nodes, get data at the highest possible rate C. For analysis, we consider a single-cell system with N as 1000. The radius of the cell and the relay cells are considered 2500 and 500 meters respectively. We considered received data rates and corresponding distances from Table I as given in [4]. We increased the number of relay nodes gradually in equation (2) to evaluate the system throughput. 3000 throughput with relay stations − pre−engineered approach Throughput without relay stations − round robin scheduling Throughput without relay stations − opportunistic scheduling
System Throughput (Kbit/s)
2500
2000
1500
1000
500
0
Fig. 5.
0
10
20
30
60 50 40 no. of relay stations
70
80
90
B. Blocking Probability We define blocking probability as the probability with which a user’s request is rejected i.e., the request is denied the permission to even enter the waiting queue. Let us first consider a system without relay nodes where each user is guaranteed a minimum data rate regardless of his location. If the server has enough bandwidth available to serve all the existing users along with the new user providing at least their guaranteed data rates to all of them, then only a new user is admitted into the server. Otherwise, the new user is requested to stay in the queue, for some time, after which, upon availability of the bandwidth the new user is admitted. We assume a queue of size M . We denote the data rate perceived by a user in the ith ring in his/her own slot as Ci Kbit/sec (i = 1, · · · , k) and the guaranteed minimum data rate provided to all the users as Cmin Kbit/sec for the entire time, where Cmin S We assume every user intends to download file of size F . Note: µS is not a constant but rather depends on the perceived data rates of S users being served in the server in time slotted mechanism. Then, the equilibrium probabilities P (j) are related to each other through the following state balance equations: µj P (j) = λP (j − 1) if 1 ≤ j ≤ S (5) µS P (j) = λP (j − 1) if S < j ≤ S + M
= P (S + M ) =
Using the above equations recursively along with the normalS+M ization condition we have, j=0 P (j) = 1. Then the steady state probabilities can be given by, j P (0) if 1 ≤ j ≤ S jλ µi i=1 P (j) = (6) j λ P (0) if S < j ≤ S + M j j−S (µS )
(
i=1
Ring k
Data rate ( Kbit/sec )
Radius (m.) (α = 4)
11 10 9 8 7 6 5 4 3 2 1
38.4 76.8 102.6 153.6 204.8 307.2 614.4 921.6 1228.8 1843.2 2457.6
2820 2370 2210 2000 1860 1680 1410 1280 1190 1070 1000
µi )
TABLE I R ATES AND R ING R ADIUS
where,
In figure 7, we plot the blocking probability against arrival rate graphs for with and without RNs as derived from (7) equations (8) and (10). As evident from the figure, blocking j=1 probability decreases with introduction of RNs. Of course, the Based on the above analysis and probability definitions, we cost of deploying the RNs is not considered here. can define P (S + M ) as the Blocking Probability (BP ). Then blocking probability of this time-slotted CDMA/HDR model can be given as,
S λj P (0) = 1 + j
S+M
λj + j−S ( j i=1 µi i=1 µi ) j=S+1 (µS )
−1
Without Relay Station With Relay Station
0.5
= P (S + M ) =
λS+M P (0) S (µS )M ( i=1 µi )
0.4
(8)
Let us now analyze the system with the relay nodes. As the relay nodes are just overlaid on the existing system model without RNs, we keep all the aspect of the previous model fixed but the service rate. As the RNs transmit at a short range with Wi-Fi like mechanism, loss due to fading and shadowing can be assumed almost negligible and data rates perceived by the users are the maximum possible for the users under the direct influence of the RNs. Then the service rate in this model can be given by, j Cmax i=1 if j ≤ S jF (9) µj = µS if j > S where, Cmax is the maximum possible data rate perceived by the users under the direct influence of the RNs and moreover,
IEEE Communications Society / WCNC 2005
Blocking Probability
BP
0.3
0.2
0.1
0 0
Fig. 7.
0.05
0.1
0.15
0.2 0.25 Arrival Rate
0.3
0.35
0.4
0.45
Blocking Probability with and without relay nodes V. S IMULATION R ESULTS
To verify our proposed model we performed simulation experiments on a UNIX based platform. We considered that the users are scattered randomly over the cell. For the relay nodes, we considered both situations - randomly placed and pre-engineered. Random placement would incur overlaps of relay cells whereas the pre-engineered would not till a certain number of relay nodes.
1712
0-7803-8966-2/05/$20.00 © 2005 IEEE
A. Throughput vs. number of relay nodes We considered N (number of active users) as 1000, radius of the cell (R) as 2500m and radius of the relay cell (r) as 500m. The throughputs are shown in figure 8. For the preengineered case, the throughput is almost the same as was obtained analytically. With increase of relay nodes, more users come under the direct influence of relay nodes. When all the users are covered by the relay nodes, the system throughput reaches its maximum value.
relay nodes does not vary much. This is because all the users are close to the base station and receive very high rate. But when the cell radius increases, more and more number of users get lesser signal strength and hence lesser data rate, resulting in decreased system throughput. With the presence of relay nodes, the far off users are virtually brought near the base station allowing them to receive higher data rate. The figure also shows that for covering a larger cell, more relay nodes would be required if the same throughput is to be expected. 3000
3000
throughput with 40 relay stations throughput with 25 relay stations throughput without relay
2500
2500
2000
2000
Throughput (Kbit/s)
System Throughput (Kbit/s)
throughput with relay stations − pre−engineered approach throughput with relay stations − random approach Throughput without relay stations − opportunistic scheduling policy Throughput without relay stations − round robin scheduling policy
1500
1000
1000
500
500
0
0 0
10
Fig. 8.
20
30
60 50 40 no. of relay stations
80
70
90
100
22 throughput with 40 relay stations throughput with 25 relay stations throughput without relay
18
1000
2500 2000 1500 Radius of the cell (meter)
3000
3500
4000
Effect of radius ratio
In this paper, we proposed augmentation of 3G cellular networks with Wi-Fi like relay nodes to provide users with high data rates regardless of their location with respect to the base station. We analytically showed how uniform coverage within a cell can be obtained by the introduction of relay nodes. We also showed how pre-engineered positioning of the RNs can provide better performance compared to random placement. We performed simulation experiments with data rates of HDR to evaluate the throughput and blocking probability of the system as well as per-user throughput for both the deployment strategies. It is shown that the use of relay stations always enhances the performance, relatively better if pre-engineered.
16
R EFERENCES
14
12
10 8 6 4 2
0
500
VI. C ONCLUSIONS
Here, we study the per-user throughput for both with and without relay nodes. We simulated with the same values for R and r as mentioned earlier but considered two different number of relay nodes - 25 and 40. The results obtained are shown in figure 9. As expected, with the increase in number of active users, the per-user throughput decreases because of resource sharing, but the rate of decrease is lesser for the case with relay nodes than without relay nodes. With increased number of relay nodes, this decreasing rate can be further reduced. 20
0
Fig. 10.
Throughput: random and pre-engineered
B. Per-user Throughput vs. number of users
Throughput per user (Kbit/s)
1500
0
200
Fig. 9.
400
600
800
1200 1000 no. of users
1400
1600
1800
2000
per-user Throughput: random deployment
C. Throughput vs. radius of cell So far, the radii of the cell and the relay nodes were considered constant. Though the absolute value is of less importance, what is decisive is the ratio of the two radii. The results obtained so far assumed a ratio of 1 : 5 (500m : 2500m). We keep the radius of the relay node constant (500m) and vary the radius of the cell to 4000m so as to get a varying radius ratio. We considered the same number of relay nodes25 and 40. The throughputs with and without relay nodes with varying cell radius are shown in figure 10. We observe that when the radius (R) is small, throughput with and without
IEEE Communications Society / WCNC 2005
1713
[1] 3GPP TR25.848 V0.5.0, “Physical Layer Aspects of UTRA High Speed Downlink Packet Access(HSDPA),” ANNEX B, TSGR1#18(01)186, Jan 18,2001. [2] G. Aggelou and R. Tafazolli, “On the relaying capacity of nextgeneration GSM cellular networks,” IEEE Personal Communications Magazine, 8(1):40-47, Feb. 2001. [3] P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushyana, S. Viterbi, “CDMA/HDR: a bandwidth efficient high speed wireless data service for nomadic users”, IEEE Communications Magazine, Volume: 38 Issue: 7, July 2000, pp. 70-77. [4] T. Bonald and A. Proutiere, “Wireless Downlink Data Channels: User Performance and Cell Dimensioning”, Proceedings of ACM Mobicom, 2003, pp. 339-352. [5] H. Hsieh and R. Sivakumar, “On Using the Ad-hoc Network Model in Wireless Packet Data Networks,”In Proceedings of ACM MobiHoc,2002. [6] L. Kleinrock and J. Silvester, “Optimum transmission radii for packet radio networks or why six is a magic number,” Proc. IEEE National Telecommunications Conference, 1978, pp. 4.3.1-4.3.5. [7] Y. Lin and Y. Hsu, “Multihop Cellular: A New Architecture for Wireless Communications,” In Proceedings of IEEE INFOCOM, volume 3, 2000. [8] H. Luo, R. Ramjeey, P. Sinha, L. Li, S. Lu, “UCAN: A Unified Cellular and AdHoc Network Architecture,” In proceedings of Mobicom, 2003. [9] H. Wu, C. Qiao, S. De and O. Tonguz, “iCAR: Integrated cellular and ad hoc relaying systems,” IEEE Journal on Selected Areas in Communications, Volume: 19, Issue: 10, pp. 2105-2115.
0-7803-8966-2/05/$20.00 © 2005 IEEE
