Combined Uplink Resource Reservation and Power Control for Relay-Based OFDMA Systems Ralf Pabst, Daniel C. Schultz, Bernhard H. Walke RWTH Aachen University, Faculty 6, Chair of Communication Networks, Germany EMail:
[email protected]
Abstract— This work presents a combined Uplink Resource Reservation and Power Control framework for Relay-Based OFDMA systems, i.e. for typical IMT-Advanced candidate systems. The framework serves to provide both fair uplink scheduling and a balanced link budget in a relay deployment. The performance of the proposed framework in a suburban cellular setup is evaluated by means of stochastic, event-driven simulation.
I. I NTRODUCTION Relay-based radio network deployment concepts have been proposed and studied in the context of candidate technologies for IMT-Advanced systems [1]. The motivation is to enable a low cost broadband network deployment by overcoming the range-limitations of broadband air interfaces, especially at high carrier frequencies above 3 GHz. Large research initiatives like the IST-WINNER project [2] consider relays to be an inherent building block of next-generation wireless systems. Fixed relays for coverage and capacity enhancement are already being standardized by IEEE802.16, Task Group “j” [3] under the term Mobile Multihop Relays (MMR). Orthogonal Frequency-Division Multiple Access (OFDMA) is the prevailing transmission scheme for next-generation Radio Access Networks. OFDMA-based systems can flexibility sub-divide the available resources in the frequency-domain. The bidirectional (UL/DL) nature of the communication in mobile and wireless networks requires establishing a balanced link budget (UL/DL cell sizes), because the cell size is determined by whichever link has the smaller range. Unbalanced transmission ranges imply resource under-utilization. Factors that influence the link budget are • different peak output powers of different station types • different noise figures / receiver gains / sensitivities • different service SINR requirements • power control constraints An additional requirement is to cope with the effects of frequency synchronization errors and - more severely - Doppler shift, multi-carrier systems with an FDMA component in the UL require an equalized reception of UL transmissions at the BS within a tolerable dynamic range [4]. From this results the need for an uplink power control scheme that takes into account the above constraints. This scheme will consequently have interrelations with the resource allocation (resource scheduling). Terminals located at the cell edge will have to concentrate their available transmission power on a small number of frequency resources. In turn, users
978-1-4244-1645-5/08/$25.00 ©2008 IEEE
2248
close to the centre of the cell at the same time may not be allowed to use all their available power and maybe also not be able to use the optimal modulation and coding if scheduled in parallel with a “bad” (i.e. cell-edge user). Separating good and bad users in the time-domain could hence contribute to an increase in system capacity, a technique which goes beyond the scope of this paper and is considered to be for further study. It is a trivial observation that for a balanced link budget, the UL power spectral density will have to be in the same order of magnitude as in DL. Eventual differences will mainly be the result of different noise figures, receiver gains/sensitivities and different SINR requirements caused by different UL/DL traffic and service characteristics. As a consequence of what was said above and owing to lower peak power in UTs, less spectral resources can be used. Figure 1 on the following page shows the effect of RAP power advantage (denoted α: “alpha”) on the max number of UL resources (denoted nU T , normalized to the number of DL resources nBS which the BS uses in parallel). The allocation of spectral resources to UL users is not only influenced by their power and channel condition but also by the amount of capacity the users have requested for their UL transmissions. Therefore, this work proposes a framework for a joint resource reservation and UL power control scheme applicable to arbitrary OFDMA based systems. Another novel aspect of the proposed scheme is that it is extended to relaybased deployments, taking into account: • •
Different UL power classes of Relay Nodes (RN) and User Terminals (UT) Strongly varying capacity requests between RNs and UTs
The remainder of this paper is organized as follows. Section IV briefly introduces the resource reservation strategies that are being compared in the simulations. Section II analytically derives limitations and requirements for the UL transmit power of the UTs and RNs. Sections III and IV describe the proposed Power Control and Resource Allocation algorithm, while Section V provides exemplary simulation results, giving a proof of concept for the Power Control algorithm and assessing the performance of the proposed resource reservation schemes. Section VI concludes the paper and gives an outlook on further work.
TABLE I U SED SYMBOLS BS (DL) maximum Transmission Power UT (UL) maximum Transmission Power BS (DL) Transmission Power per RAU UT (UL) Transmission Power per RAU Number of RAUs used in parallel by the BS Number of RAUs used in parallel by the UT BS Antenna Gain UT Antenna Gain Distance Pathloss DL Interference per RAU UL Interference per RAU BS Noise Figure UT Noise Figure Noise Power Spectral Density Bandwidth of RAU SINR
Δ γ = 0 dB Δ γ = 3 dB Δ γ = 6 dB
1.5 nUT / nBS
PBS,M AX PU T,M AX PBS PU T nBS nU T GBS GU T d L(d) IDL IU L FBS FU T ΦN ΔfRAU γ
2
1
0.5
0
1
2
3
4 5 6 7 8 9 10 11 12 13 14 15 BS Power advantage α [dB]
Fig. 1. Ratio of resources usable in UL vs. DL under the assumption of a 2 dB difference between RAP and UT noise figures 30 25
This section analytically derives the uplink transmit power budget for cell-edge user. The analysis can be easily extended to the multi-hop case which involves relay nodes by exchanging the UT indices with RN indices. Table I lists the symbol definitions used throughout this paper. The transmission Power Advantage α of the BS as compared to the UT is:
20
PBS,M AX α= PU T,M AX
(1)
Downlink SINR of one RAU experienced by a UT at distance d from the BS: γDL (d) =
PBS GBS L(d) GU T IDL + ΔfRAU ΦN FU T
(3)
Assuming a noise-dominated scenario for simplicity, we can neglect the interference: IDL = 0,
IU L = 0
(4)
Taking into account a certain asymmetry of the service and the related QoS between UL and DL, this shall be expressed as Δγ. The requirement for a balanced link budget is then γDL (d) = γU L (d) · Δγ
(5)
Equations (2), (3), (4) and (5) then yield: FU T PBS = Δγ PU T FBS
15
n
UT,i
/n
Δ Rx = 0 dB Δ Rx = 3 dB Δ Rx = 6 dB Δ Rx = 9 dB
10 5 0 0
20
40
60
80 100 120 140 Distance of UTi from BS [m]
160
180
200
Fig. 2. Max. number of RAUs in parallel for a user at distance di , the user with the worst path loss is located at dw = 150 m.
equal power to all RAUs. The maximum power stations can use to serve a single RAU in order to not exceed their maximum power constraints is then:
(2)
Uplink SINR of one RAU experienced by the BS from the same UT at distance d: PU T GBS L(d) GU T γU L (d) = IU L + ΔfRAU ΦN FBS
UT,w
II. T RANSMIT P OWER B UDGET
(6)
We allow that the BS serves nBS RAUs in parallel (on different OFDMA subchannels and/or different spatial layers). For an equalized OFDM signal, we assume the BS to allocate
2249
PBS,M AX ≤ PBS nBS ,
resp.
PU T,M AX ≤ PU T nU T (7)
Using this assumption together with (6) and (1) we find the maximum fraction of the number of RAUs available at the BS (nBS ) that the UT may use in parallel (nU T ) is: Δγ FBS nU T ≤ nBS α FU T
(8)
As stated above, Δγ is a measure for the envisaged asymmetry between DL and UL Service SINR Requirements and F denotes the Noise Figures of BS and UT respectively. Figure 1 plots this for the assumption of a 2 dB difference in BS and UT noise figures. The x-Axis varies the BS transmit power advantage α while the parameter is the asymmetry factor Δγ. The above consideration is only valid for UTs at the cell border. III. P OWER A LLOCATION A LGORITHM The power allocation considerations in the previous section concern a user at the cell border. Another constraint on UL OFDMA transmissions is that all users’ transmission are required to reach the BS within a certain dynamic range, further
referred to as ΔRx. Not exceeding this dynamic range is vital to preserve the orthogonality between signals on adjacent subchannels. In a cell with I users at individual locations di , this constraint can be formulated as follows: Rxmin · ΔRx = Rxmax
(9)
3) taking into account their respective pathloss, set the number of parallel resources and the TX power per spectral resource for all other UTs to yield that target RX power (plus a dynamic range ΔRx), according to (16). IV. S CHEDULING AND R ESOURCE R ESERVATION
A fair scheduling algorithm is further considered that takes into account for i = 1 . . . I Rxmin ≤ PU T,i GBS L(di ) GU T ≤ Rxmax • The constraints on parallel resources and the allowable (10) power limits as identified by the Power Control algorithm It is a fair assumption that the lower end of this range will from the previous section be represented by the user with the highest pathloss (located • The resources requested by UL users (i.e. both RNs at the cell edge or in a heavily shadowed location, we further and UTs). The resource allocation strategy calculates the denote this user with the index w). In (8) we have determined overall sum of resources requested by the individual UL the relative number of RAUs nU T,w /nBS that this user may stations and grants each station a share that corresponds use in parallel within its power budget and from (7) we obtain to the percentage of overall resources that the respective the uplink received power per RAU from this user. station has requested. In overload situations, this leads to a graceful degradation, i.e. if the overall demand can not PU T,M AX be satisfied, each users’ granted resources are reduced by GBS L(dw ) GU T (11) Rxw = Rxmin = nU T,w an equal percentage. The proposed scheme currently encompasses two reOwing to the equalization constraint in (9) and (10), other source request/reservation strategies users with lower pathloss will allocate less power per RAU. They can consequently afford to transmit on a higher number – single-hop reservation: users inform the serving staof parallel RAUs than user w. tion about their own buffer fill levels – multi-hop reservation: users inform the serving stafor i = 1 . . . I, i = w (12) PU T,i ≤ PU T,w tion about their own buffer fill levels plus the reported fill-levels of the subsequent tier’s stations Under the above constraint, Rxi has to follow this inequalThe section on simulation results compares the performance ity: of the two reservation strategies under traffic congestion in the Relay Enhanced Cell (REC). PU T,M AX GBS L(dw ) GU T Rxi ≤ ΔRx · Rxw = ΔRx · V. S IMULATION R ESULTS nU T,w (13) The simulation results shown in this paper have been obtained in the same scenario setup as the one described in (14) [5]. A 7-cell hexagonal setup has been simulated where only Rxi = PU T,i GBS L(di ) GU T the users in the center cell were used for evaluation. Each cell was enhanced by 3 RNs, also placed according to a regular L(d)w PU T,M AX · (15) hexagonal grid, resulting in a total of 7 Relay Enhanced Cells =⇒ PU T,i ≤ ΔRx L(d)i nU T,w (REC). Or, to determine the relative number of parallel RAUs that User data traffic has been modeled according to a per-user user i can use: Poisson arrival process (neg. exp. Distributed packet interarrival times). All simulations were run with a fixed number of 1 L(d)i nU T,i ≥ (16) 44 Users per REC. Scaling of the overall traffic load per cell nU T,w ΔRx L(d)w has been performed by scaling each user’ share of the traffic. Not surprisingly the advantage in the number of RAUs for Example: At 50 Mbit/s cell load, each user has an individual user i comprises the pathloss advantage this user L(d)i /L(d)w offer of 50e3 / 44 = 1.136 MBit/s. At a fixed packet size has over the ’worst’ user w and the allowable dynamic range. of 128 byte, this corresponds to an average of 1110 packets Figure 1 on the preceding page illustrates that. per second and user. User terminals within the area of a REC The power control algorithm takes into account the above are evenly distributed in circular areas around each RAP (see considerations by applying the following procedure: Fig. 1). As an exemplary distribution, the 44 UTs in the cell 1) Determine the UT in the (sub-)cell which has the highest were divided into 26 UTs immediately connected to the BS path loss to the BS or RN and 18 UTs connected via the 3 RNs (6 each). In the single2) According to (8), determine the maximum number of hop comparison case, all 44 Users were connected to the BS parallel resources this UT can use (and estimate the immediately but remained in the same positions as in the resulting RX power at the BS/RN). multi-hop simulations. In the case of bottlenecks, all nodes can
2250
1
100
0.9
90
0.8
80
1−Hop Conns., trad. res. 2−Hop Conns., trad. res.
Sustained UL cell traffic [MBit/s]
All Conns., trad. res.
0.7
P(X
