- Research Article
- Open Access
Resource Allocation with EGOS Constraint in Multicell OFDMA Communication Systems: Combating Intercell Interference
© Husheng Li. 2010
- Received: 23 October 2009
- Accepted: 22 June 2010
- Published: 11 July 2010
Resource (power and bandwidth) allocation is an important issue in orthogonal frequency division multiple access (OFDMA) systems. For multicell systems, the interference across different cells makes the optimization of resource allocation difficult. For finite systems, a constraint on the rise over thermal (ROT) is placed to alleviate the intercell interference. A hybrid scheme with equal receive power and peak transmit power is shown to be optimal for the ROT constrained case. Large system analysis is applied for multi-cell OFDMA systems with the fairness constraint of equal grade of service (EGOS). An interference function is defined to model the intercell interference. Variational analysis is used to compute the optimal profile of transmit power and bandwidth. The optimal resource allocation is then computed using numerical simulations.
- Resource Allocation
- Peak Power
- Power Allocation
- Channel Gain
- Orthogonal Frequency Division Multiple Access
In contrast to the analysis on single-cell OFDMA systems, the study on multi-cell OFDMA systems is much less [7–12]. Most studies focus on the collaboration across different cells, for example, using noncooperative game theory. However, in practical systems, it incurs much overhead to maintain frequent coordinations across different cells. In [13, 14], the power control of multiple cells is studied without the coordination among multiple cells. Different from this paper, they focus on the down link and do not consider the allocation of bandwidth.
Therefore, in this paper, we studythe situation where there is no explicit cooperation across different cells while each user's resource (power and bandwidth) is stationarily allocated with the awareness of inter-cell interference. It can be applied in practical OFDMA systems as an open-loop control strategy and provides reasonably good initial values for inter-cell coordination-based dynamic algorithms. The key difficulty is how to model and handle the inter-cell interference. One approach is to place a constraint on the total received power at the base station, which is widely used in industry. Then, the problem of resource allocation is converted into a constrained optimization. Such an approach is applicable in finite systems. An alternative approach is to apply the large system analysis, that is, the number of active users tends to infinity, to alleviate this difficulty by defining an interference function. Variational analysis is then used to obtain a functional equation characterizing the optimal transmit power profile. Note that a fairness criterion is necessary for the resource allocation. For simplicity, we used the criterion equal grade of service (EGOS), that is, the throughput of each user within the same cell should be the same. It is straightforward to extend the framework of analysis proposed in this paper to other criteria like proportional fairness.
The remainder of this paper is organized as follows. The system model is introduced in Section 2. The resource allocation in finite systems is discussed in Section 3. The optimal resource allocation is derived for large systems in Section 4. Numerical results and conclusions are provided in Sections 5 and 6, respectively.
Suppose that each cell has neighboring cells. We assume that there are access terminals (ATs), namely users, and one access point (AP), namely base station, in each cell. It is straightforward to extend the discussion to the case of different numbers of ATs in different cells. We denote by the channel power gain of AT in cell , whose cumulative distribution is denoted by . We assume that the channel state information is perfectly known, which can be achieved by letting ATs sending pilots for channel estimation.
The communication of ATs is confined by peak transmit power and total bandwidth . We denote by and the transmit power and bandwidth allocated to AT at cell , which satisfies and . We assume that the bandwidth is sufficiently large such that we can consider the allocated bandwidth as a continuous real number, thus simplifying the analysis.
We denote by the transmission rate of AT at cell and assume the EGOS fairness within the same cell, that is, .
We assume that the transmitters carry out frequency hopping in every OFDM symbol period and the hopping sequence is pseudo random. Therefore, the inter-cell interference is averaged and can be considered as Gaussian noise. Note that, if opportunistic subcarrier allocation is used, the assumption random frequency hopping is nolonger valid. Then, the inter-cell interference is not averaged and thus becomes frequency selective (e.g., an edge user will cause more interference to the neighboring base station in the subcarriers allocated to this user). In such a scenario, the analysis becomes more complicated since the interference power spectral density is a function of the resource allocation. A new approach is needed for finding the optimal or near-optimal scheduling policy, which is beyond the scope of this paper.
- whereand are the power spectral densities (PSDs) of additive white Gaussian noise and interference. Note that the interference PSD is identical for all ATs with the same cell due to the assumption of frequency hopping. For suboptimal coding schemes, we can use the following expression to approximate the reliable transmission rate:
where is the gap to Shannon capacity .
Each cell has no information about the ATs of other cells, for example, locations, channel gains, and transmit powers. Therefore, the resource allocation is carried out within individual cells separately.
where is the total number of cells under consideration. Essentially, the optimization maximizes the total throughput of all cells, under the constraint of limited transmit power and limited total bandwidth. It is difficult to solve the optimization problem via explicit analysis for finite systems. Moreover, this is not a convex optimization problem since the equality constraint is not affine. Therefore, we propose a heuristic approach for finite systems in the next section. In Section 4, we will use large system analysis to alleviate the key difficulties in finite systems.
For finite systems, that is, , it is computationally prohibitive to carry out the precise optimization across different cells, particularly when is large. Moreover, it requires coordinations cross different cells, which contradicts the assumption of no explicit inter-cell coordinations. Therefore, we apply a heuristic approach for the resource allocation in each cell without the information of other cells. For simplicity of notation, we ignore the cell index, , in all notations since we focus on only one cell.
The heuristic approach is to confine the total received power at AP (Rise Over Thermal, ROT) (the standard definition for ROT should be the ratio of the total received power over the thermal noise power. For simplicity, we define ROT as the total received power since the noise power is fixed and is known), which is based on the intuition that ROT is roughly proportional to the transmit powers of ATs within the desired cell and also positively correlated with the interference to other cells. Note that the metric of ROT is used to measure the congestion level of the cell in code-division multiple access (CDMA) systems . Meanwhile, there is also requirement for ROT from the hardware viewpoint; if the ROT is too high, that is, the total received power is too high, weak signals may be lost due to the limited number of levels of analog-digital converter (ADC).
Otherwise the ROT constraint is useless.
First, we need to study the feasibility of the optimization, which is assured by the following lemma. The proof is given in Appendix A.
There exist feasible resource allocation schemes satisfying the constraints.
The following lemma shows that, if we can find a new resource allocation scheme (not necessarily satisfying the EGOS constraint) such that the transmission rates are improved or remain unchanged for every AT, compared to an old EGOS allocation scheme, we can always find an EGOS resource allocation scheme that is better than the old one. The proof is given in Appendix B.
For a -AT cell, suppose that a resource allocation and satisfies the EGOS constraint. If there exist a different resource allocation scheme and such that there exists a such that and for ( and are not necessarily equal), we can always find an EGOS allocation scheme and such that .
Intuitively, the resource allocation should fully utilize the budget of total bandwidth and ROT. This intuition is proved rigorously in the following lemma. The proof is given in Appendix C.
For an optimal resource allocation, the ROT and total bandwidth constraints should be equalities.
For exploiting the property of the solution to the optimization problem in (4), we show the following lemma, which states that when there are only two ATs, the AT with a better channel condition should use a larger receive power. The proof is given in Appendix D.
For a two-AT cell ( ) and constrained ROT case, suppose that an allocation and satisfies , and . Then, we can always find a better allocation scheme yielding higher throughout and satisfying the EGOS and ROT constraints.
Based on Lemma 4, we obtain the following proposition, which discloses a necessary condition of optimal ROT constraint-based resource allocation. The proof is given in Appendix E.
all ATs having channel gains smaller than or equal to should transmit with peak power;
all users with channel gains larger than should transmit with equal receive power .
Due to Proposition 1, the optimization is simplified to the task of finding the two optimal parameters, namely and , which can be determined by using the following steps (without loss of generality, we assume that ):
Step :set .
If there exists an such that , go to Step .
Then, record the EGOS transmission rate . Let and go to Step .
Step :Compare and choose the scheme yielding the highest EGOS transmission rate.
Note that the quantity can also be optimized using the algorithm: different feasible can be evaluated using the above algorithm; then the achieving the highest EGOS rate should be adopted.
In this section, we study the resource allocation for large systems. We first explain the large system analysis and apply variational analysis to obtain a condition for optimal resource allocation. Then, we propose an algorithm for computing it and discuss the optimality of all peak power scheme.
4.1. Large System
As mentioned in Section 1, the difficulty of the optimal resource allocation is that the interference PSD at one cell is determined by other cells and thus cross-cell optimization is required. However, there is no explicit information exchange between neighboring cells. Mathematically, this difficulty can be alleviated by large system analysis, namely letting the number of ATs, , and the total bandwidth, , tend to infinity while keeping their radio , which means the average bandwidth per AT, a constant. When the system size tends to infinity, the interference PSD will converge to a constant, which is equal for all cells.
since there is no explicit coordination across different cells, the resource allocation to an AT is determined by only the channel gain of the AT to the serving AP, which can be denoted by and . As , the resource allocation can be considered as two functions of the channel gain,
- (ii)for evaluating the interference of an AT to the AP of a neighboring cell, we define the corresponding interference function as
where is the channel power gain to its serving AP and is the channel power gain to the neighboring AP. It is impossible to derive an explicit expression for the interference function. However, it can be approximated by simulations or field experiment results. In Section 5, we will evaluate the interference function by adopting a certain wireless channel model.
4.2. Variational Analysis
Change the power allocation function by a sufficiently small .
We adjust the bandwidth allocation such that all ATs still keep the original EGOS rate corresponding to in the following way: for ATs with , they donate some bandwidth to a "bandwidth bank" (illustrated in Figure 2) to decrease their transmission rates; for ATs with , they borrow some bandwidth from the "bandwidth bank" to improve their transmission rates.
Then, we check the net income of the "bandwidth bank": if it is positive, the remaining bandwidth can be distributed to all ATs to improve the EGOS rate, thus finding a better EGOS resource allocation scheme.
The details of the above steps will be discussed in the remainder of this subsection.
where is the characteristic function, then we can obtain (it is easy to verify by substituting (17) into (15)), that is, we can distribute the spared bandwidth to all ATs and improve the EGOS throughput.
Based on the above analysis, we obtain the following lemma (we coin the condition " condition").
for all , ;
for any such that , .
The second item implies that, if , .
In Lemma 5, we derived only a necessary condition for the optimal resource allocation. The following proposition shows that these conditions are also sufficient for a locally optimal resource allocation. (A locally optimal resource allocation means that all other resource allocation schemes within a neighborhood of it achieve worse or equal EGOS transmission rate.) The proof is given in Appendix F.
A resource allocation is locally optimal if and only if the condition holds.
Proposition 2 also provides an efficient approach to compute the optimal resource allocation scheme, which can be iteratively done using the following steps.
Step :discretize the channel gains and approximate the functions and by using a finite number of s. Initialize all transmit powers and compute the corresponding bandwidth allocation that assures EGOS constraint.
Step :Compute according to (16) and the corresponding power change by using (17) and a sufficiently small . Note that all integrals are approximated by discrete summations.
Step :Update the power allocation to . Check the stopping rule (either the maximum number of iterations or the difference between the resource allocations of successive iterations). If not stopping, go back to Step .
4.4. Peak Power or Not
It is easy to verify that decreases as is increased.
Thermal limited ( ): the right-hand side (RHS) in (18) is larger than the left-hand side (LHS). Therefore, the inequality in (18) holds and all ATs should transmit in peak power. This also coincides with our intuition.
Broadband ( is sufficiently large): the inequality in (18) also holds and all ATs should transmit in peak power.
Notice that integrating over the probability measure yields 1 on both sides. Then, there exists a such that the inequality does not hold, unless , . Therefore, the all peak power allocation scheme is not optimal.
When changes sufficiently rapidly, the LHS of (21) is larger for some , which implies that the all peak power scheme is suboptimal.
By integrating over on both sides, we obtain , which is impossible. Therefore, we obtain the following proposition.
In any locally optimal power allocation, the proportion of ATs transmitting with peak power is nonzero.
We now illustrate the analytical results in this paper via simulations. For comparison, we also simulated the performance of two alternative resource allocation schemes, namely all peak power and equal receive power schemes.
5.1. Interference Function
where , and are i.i.d. Gaussian random variables (expectation zero and variance 8.9 dB) and . The physical meaning of (23) is as follows: the shadow fading between an AP and an AT consists of three independent components; represents the shadowing effect around AP is common for AP ; stands for the shadowing effect around AT ; similarly, represents the shadowing effect around AT is common for AP . The requirement normalizes the shadowing effect. For simplicity, we assume . This formulation is widely used in numerical simulations in industry.
5.2. Optimal Transmit Power
peak power: every AT transmits with peak power (we assume that peak power is 200?mW);
equal receive power: every AT transmits with equal received power such that the AT having the lowest channel gain transmits with peak power .
We also computed the ROT-based resource allocation using the algorithm proposed in Section 3. The computation results in the scheme of equal receive power.
5.3. EGOS Transmission Rate
We have discussed the resource allocation (power and bandwidth) of multi-cell OFDMA communication systems under the fairness constraint of EGOS. For finite systems, we have applied the heuristic ROT constraint to derive the resource allocation scheme. On assuming sufficiently many users within each cell, we have applied the large system analysis and variational analysis to obtain the optimal power and bandwidth profiles. To model the intercell interference, we have defined an interference function which is obtained from numerical simulations. We have used numerical simulations to demonstrate the effectiveness of the proposed algorithm.
A. Proof of Lemma 1
It is easy to find a resource allocation satisfying the constraints of peak power, total bandwidth, and ROT. The only thing we need to check is the EGOS constraint.
We assume that when , when are fixed, there exists a unique such that and is a continuous function of .
Now, we consider the case . Again, we fix . We range the bandwidth allocated to AT1, , from 0 to and the corresponding total bandwidth allocated to the remaining ATs, , is ranged from to 0. Similar to the 2-AT case, ranges from 0 to and the EGOS transmission rate for the remaining ATs ranges from to 0. Due to the continuity and monotonicity of and (due to the induction assumption), there exists a unique such that . It is easy to verify the continuity of using the same argument as the 2-AT case. This concludes the proof.
B. Proof of Lemma 2
We fix and change only the bandwidth allocation to improve the performance. Since , we can find a sufficiently small such that when , we have . Then, the spared bandwidth can be allocated to the remaining ATs. Using a similar argument to that in Lemma 1, we can find an allocation such that , , and the equal rate is a continuous and monotonically increasing function of . When ranging from 0 to , ranges from to 0 while ( ) ranges from being less than to being larger than . Due to the continuity, we can find a unique such that EGOS is satisfied and . This concludes the proof.
C. Proof of Lemma 3
Suppose that the ROT constraint is an inequality for an optimal resource allocation. Then, we can increase the transmit power of an AT, for example, AT , whose current transmit power is below the peak power (we can always find such an AT due to the assumption ). Then, the transmission rate of AT is increased while all other ATs remain the same rate. By applying Lemma 2, we can always find a better EGOS allocation scheme, which contradicts the optimality.
Suppose that the total bandwidth constraint is an inequality. Then, we can allocate the unused bandwith to an arbitrary AT. A new EGOS allocation scheme that achieves a better performance exists by applying Lemma 2. The contradiction concludes the proof.
D. Proof of Lemma 4
Note that can be increased since . We also add a sufficiently small bandwidth to AT 1 and reduce from AT 2. Notice that the changes in power and bandwidth still satisfy the bandwidth, ROT and peak power constraints.
which are obtained simply by Taylor's expansion.
Applying the fact that and , it is easy to check that and . Therefore, it is always possible to find sufficiently small and satisfying (D.5) and yielding better performances for both ATs. The proof is concluded by applying Lemma 2.
E. Proof of Proposition 1
First, we prove that ATs not transmitting with peak power must transmit with equal receive power. Suppose that two such ATs have different channel gains and transmit with different receive power. Applying Lemma 4, we can find a resource allocation scheme for these two ATs such that the transmission rates of these two ATs are increased while keeping the total ROT and bandwidth unchanged. By applying Lemma 2, we can always find a better EGOS resource allocation scheme, which contradicts the assumption of optimality.
Then, we prove that ATs transmitting with peak power must have smaller channel gains than ATs not transmitting with peak power (as we have shown, these ATs must transmit with equal receive power). Suppose that, in the optimal scheme, AT transmits with peak power, AT 's transmit power is less than and . Then, the receive power of user is higher than that of user . By applying Lemma 4, we can change the power and bandwidth allocation of these two ATs to improve their performance without violating the constraints. The conclusion is obtained by applying Lemma 2.
This concludes the proof.
F. Proof of Proposition 2
The necessity has been established in the proof of Lemma 5. Therefore, we focus on the proof of sufficiency. Suppose that a resource allocation scheme and , yielding EGOS transmission rate , satisfies the condition and there exists a better resource allocation scheme in each neighborhood. Then, for a sufficiently small neighborhood, we denote a better resource allocation scheme by and , which yields higher EGOS transmission rate . We define . Now, we consider two approaches to recover the original EGOS rate when the power allocation is changed to and find contradiction.
(i)Change and to and to achieve higher EGOS rate . Then, each AT can discard some bandwidth such that the EGOS rate is decreased from to . Via this approach, we obtain the total spared bandwidth .
(ii)Change to and then change to recover the EGOS rate . Applying Lemma 5, we obtain that the total change of bandwidth is negative.
Notice that, in both approaches, the power allocations and EGOS rates are changed to and . Due to the bijective mapping between power allocation and bandwidth conditioned on a fixed EGOS rate (this can be easily shown by verifying that the transmission rate is a rigorous monotonically increasing function in both power and bandwidth), the final total used bandwidth should be the same in both approaches, which implies . This contradicts the fact that and concludes the proof.
This work was supported by the National Science Foundation under Grants CCF-0830451 and ECCS-0901425.
- Choi KW, Jeon WS, Jeong DG: Resource allocation in OFDMA wireless communications systems supporting multimedia services. IEEE/ACM Transactions on Networking 2009, 17(3):926-935.View ArticleGoogle Scholar
- Gotsis AG, Constantinou P: Adaptive single-cell OFDMA resource allocation for heterogeneous data traffic. Proceedings of the 4th IEEE International Conference on Wireless and Mobile Computing, Networking and Communication (WiMob '08), October 2008 96-103.Google Scholar
- Han Z, Ray Liu KJ: Resource Allocation for Wireless Networks. Cambridge University Press, Cambridge, UK; 2008.View ArticleGoogle Scholar
- Wong CY, Cheng RS, Letaief KB, Murch RD: Multiuser OFDM with adaptive subcarrier, bit, and power allocation. IEEE Journal on Selected Areas in Communications 1999, 17(10):1747-1758. 10.1109/49.793310View ArticleGoogle Scholar
- Yin H, Liu H: An efficient multiuser loading algorithm for OFDM-based broadband wireless systems. Proceedings of the IEEE Global Communications Conference (GLOBECOM '00), 2000, San Francisco, Calif, USA 1: 103-107.Google Scholar
- Yu W, Cioffi JM: FDMA capacity of Gaussian multiple-access channels with ISI. IEEE Transactions on Communications 2002, 50(1):102-111. 10.1109/26.975766MathSciNetView ArticleGoogle Scholar
- Gesbert D, Kiani SG, Gjendemsj A, Ien GE: Adaptation, coordination, and distributed resource allocation in interference-limited wireless networks. Proceedings of the IEEE 2007, 95(12):2393-2409.View ArticleGoogle Scholar
- Li G, Liu H: Downlink dynamic resource allocation for multi-cell OFDMA system. Proceedings of the 37th Asilomar Conference on Signals, Systems and Computers (ACSSC '03), November 2003, Pacific Grove, Calif, USA 517-521.Google Scholar
- Rahman M, Yanikomeroglu H: Multicell downlink OFDM subchannel allocations using dynamic intercell coordination. Proceedings of the 50th Annual IEEE Global Telecommunications Conference (GLOBECOM '07), November 2007, Washington, DC, USA 5220-5225.Google Scholar
- Son H, Lee S: Multi-cell communications for OFDM-based asynchronous networks over multi-cell environments. Wireless Networks 2009, 15(7):917-930. 10.1007/s11276-007-0084-zView ArticleGoogle Scholar
- Wang L, Niu Z: Adaptive power control in multi-cell OFDM systems: a noncooperative game with power unit based utility. IEICE Transactions on Communications 2006, E89-B(6):1951-1954. 10.1093/ietcom/e89-b.6.1951View ArticleGoogle Scholar
- Han Z, Ji Z, Ray Liu KJ: Power minimization for multi-cell OFDM networks using distributed non-cooperative game approach. Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM '04), December 2004, Dallas, Tex, USA 3742-3747.Google Scholar
- Kiani SG, Øien GE, Gesbert D: Maximizing multicell capacity using distributed power allocation and scheduling. Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC '07), March 2007, Kowloon, Hong Kong 1692-1696.Google Scholar
- Kiani SG, Gesbert D: Optimal and distributed scheduling for multicell capacity maximization. IEEE Transactions on Wireless Communications 2008, 7(1):288-297.View ArticleGoogle Scholar
- Cioffi JM: A multicarrier primer. ANSI Contribution T1E1.4/91-157, November 1991Google Scholar
- Viterbi AJ: CDMA: Principles of Spread Spectrum Communications. Addison-Wesley, Reading, Mass, USA; 1995.MATHGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.