Dynamic Hybrid Duplex for Rate Maximization in OFDMA. Sangmin Oh and Chae Y. Lee

Transcription

1 Dynamic Hybri Duplex for Rate Maximization in OFDMA Sangmin Oh an Chae Y. Lee Dept. of Inustrial Engineering, KAIST, Kusung Dong, Taeon, Korea Tel: , FAX: {sangmin.oh, Abstract Orthogonal frequency ivision multiple access (OFDMA) is consiere as one of the technologies for 4 th generation (4G) ue to its robustness against multi-path environment. OFDMA also has the avantage of flexibility in resource allocation, which cannot be supporte by existing uplex schemes. In this paper, we propose ynamic hybri uplex (DHD) which will enhance the efficiency an flexibility of the OFDMA system. To establish the framework of resource allocation for the DHD, we formulate DHD resource allocation problem (DRAP) which maximizes the total users ata rate uner power, rate, an subcarrier allocation constraints. A heuristic algorithm is evelope to solve the problem. The algorithm fins the best moe an the best amount of resources allocate to ownlink an uplink. Simulation is performe with five scenarios to evaluate the DHD. Its results show that the propose DHD outperforms other uplex schemes in various environments. Keywors OFDMA, hybri uplex, ynamic resource allocation, flexible resource allocation 1. Introuction After the successful evelopment of 2 n generation wireless cellular communication systems, researchers have stuie 3 r generation communication for many years. Recently, the focus of the technology is moving forwar to beyon 3 r generation or 4G. 4G has many requirements an challenges: high ata rate, hanoff, quality of service, low cost, traffic asymmetry an so on [1]. Among these, one noticeable requirement is very high ata rate over a wireless channel environment. To satisfy this requirement, 4G is expecte to use broaban [1], which may be vulnerable in a wireless environment. Multi-path faing generates frequency selective faing an inter-symbol interference (ISI). Frequency selective faing makes specific frequency unergo eep faing, which ecreases efficiency of the system. ISI limits symbol rate an reuces throughput. 1

2 Among various multiple access technologies, orthogonal frequency ivision multiple access (OFDMA) is regare as a goo caniate technology because OFDMA is robust against multi-path environment [2]. Its narrow-ban subcarrier overcomes frequency selective faing an the insertion of guar interval reuces the effect of ISI. In aition to robustness against multi-path effect, OFDMA systems have the avantage of flexibility in resource allocation. In OFDMA, controllable resources are subcarriers, symbols an power, an they can be allocate very freely to users compare to other systems. However, espite its flexibility, OFDMA cannot be properly supporte by existing uplex schemes. In this paper, we propose ynamic hybri uplex (DHD) to enhance the flexibility an efficiency of OFDMA for ata transmission. It is a cross layer approach that consiers the user s traffic an channel status to allocate resources an select a proper uplex moe. Two soli uplex schemes, frequency ivision uplex (FDD) an time ivision uplex (TDD) [2, 3], are wiely implemente in many communication systems. However, FDD has the problem of inefficiency in supporting asymmetric ownlink (DL) an uplink (UP) traffic ue to its fixe banwith allocation. TDD is spotlighte because of its free slot allocation by which TDD system can be aapte to traffic asymmetry. However, TDD uses all subcarriers uring a given time perio. Thus, even ba channels are allocate to increase throughput. DHD is expecte to overcome the inefficiency of the two uplex schemes by flexibly allocating resources an properly selecting the right moe. Many research results are available for resource allocation in OFDMA systems [4 15]. In OFDMA, the channel status of each user on one subcarrier can be ifferent. Thus, the system efficiency can be increase by allocating a subcarrier to the user with goo channel status. It is the concept of multiuser iversity which is an important issue of resource allocation in OFDMA systems. Many existing stuies are base on multiuser iversity [4-15]. They can be categorize accoring to its obectives: power minimization or rate maximization. In power minimization, total transmission power is minimize uner rate constraints [4, 6-8]. In rate maximization, minimum of the user s ata rate or the total users rate is maximize uner power constraint [5, 7, 9, 11-15]. Especially, when the total users rate is maximize in DL, optimal allocation can be obtaine by a two-step algorithm: the first step is subcarrier assignment employing best user selection an the secon step is power allocation employing water-filling metho, where best user selection means that for a subcarrier only a single user who has the best channel gain is selecte [9, 11]. For the UL case, a two-step heuristic is suggeste by employing marginal user selection with which a subcarrier is allocate to 2

3 the user who has maximal marginal rate [11], an Nash bargaining solution is consiere [13]. Previous research on OFDMA can also be classifie accoring to the DL an/or UL. Most of them consier only DL [5, 7-9, 12, 15]. Pietrzyk an Janssen [6] an Kim et al. [11] consier both DL an UL. However, they o not consier how to istribute total resources to DL an UL. In this paper, we primarily scope how to istribute resources to DL an UL in OFDMA systems. With the suggestion of the DHD, we formulate resource allocation problem in OFDMA/DHD mathematically. The problem is to maximize the total users ata rate uner power an rate constraints an the restriction of allocating a subcarrier to one DL (or UL) user. We propose a heuristic to solve the problem efficiently. The heuristic is an iterative search algorithm which fins the best moe an the best amount of resources to DL an UL. The remainer of this paper is organize as follows. In Section 2, we propose DHD an iscuss its system requirements an avantages. In Section 3, we moel an formulate resource allocation in OFDMA/DHD. In Section 4, we suggest an iterative search algorithm to solve the DHD resource allocation problem (DRAP). Simulation results are shown in Section 5 with conclusion in Section Dynamic Hybri Duplex in OFDMA In OFDMA, subcarriers, symbols, an power are controllable resources. To allocate the resources flexibly to DL an UL users, we propose ynamic hybri uplex (DHD). Features, system requirements, an avantages of DHD are iscusse in this section. 2.1 Features of Dynamic Hybri Duplex DHD has the following three features as iscusse below: (1) changeable uplex moe, (2) flexible resource allocation, an (3) use of user information. (1) Changeable uplex moe DHD has three moes for uplex: FDD, reverse FDD (rfdd), an TDD moe. Reverse FDD moe is the case where the frequency bans for DL an UL are alternate compare to FDD moe. In a cellular system, DL normally employs higher frequency ban than UL. Therefore, we regar this normal situation as FDD moe. Figure 1 shows three moes of DHD. DHD can change the moe every frame. (2) Flexible resource allocation Traitionally, TDD employs flexible slot allocation while FDD employs fixe frequency ban allocation. However, in DHD, flexibility is also given to FDD an 3

4 rfdd moe. Banwith can be flexibly partitione to DL an UL epening on the system parameters or the channel status. We can show that banwith allocation for DL an UL of the FDD (or rfdd) moe are changing per frame in Figure 1. (3) Use of user information DHD takes into account users ata traffic an channel status information. Base on the information, DHD selects the moe to employ at each frame. It also ecies the require DL an UL banwiths for FDD (or rfdd) moe an the time perios for TDD moe to maximize system performance by reflecting the user information. 2.2 System Requirement First, to apply DHD, the transmitters an receivers of base stations (BSs) an mobile stations (MSs) have to be equippe with the capability to change corresponing frequency bans in real time. In OFDMA, it can be implemente by the change of the number of allocate subcarriers. However, this implementation cannot be provie easily in normal OFDMA, since the inverse fast Fourier transform (IFFT) an the fast Fourier transform (FFT), which are the core parts of the OFDMA system, require that the number of allocate subcarriers shoul be the integer power of two. To eal with this problem, we moify the OFDMA system such that the number of inputs to the IFFT (or the FFT) is the integer power of two even though the number of allocate subcarriers is not. To o that, we use total frequency ban for the IFFT (or the FFT) an allocate zero power to the unavailable subcarriers. For example, as for FDD moe in a BS, ata for DL an zeros for UL are inputs of IFFT an zeros for DL an ata for UL are outputs of FFT. Refer to [16] for more etails about IFFT an FFT. Secon, DHD requires the capability to change the moe in real time. Note that FDD an rfdd moes are easily implemente by total frequency ban usage an aforementione zero aing technique, since allocate frequency bans of rfdd moe are simply alternate as compare to FDD moe. However, in the case of TDD moe, time perios for DL an UL are separate an timing control is neee to implement the moe. To support the three moes efficiently, we a a system controller to the system as shown in Figure 2. The system controller realizes the change of moe in the system by controlling both subcarrier allocation an its allocation timing. Note that there is no elay in switching the moe in our approach. The only overhea is control signals for allocation an timing. The thir problem is how to obtain users channel information. In OFDMA, channel estimation is necessary to obtain the channel gain for various purposes. For example, the gain is use for resource allocation in meium access control (MAC) layer an for 4

5 transmission power control in physical layer. When TDD is employe as a uplex scheme in a normal OFDM system, channel information can be easily obtaine by reciprocity [16]. To obtain channel information, a MS receives a signal from a BS an estimates the channel status. This estimation can be applie to the system irectly since the same frequency ban is employe for DL an UL. On the other han, when FDD is employe, channel information shoul be feebacke by means of other link, since the frequency bans for DL an UL are ifferent. Channel information can also be obtaine by estimating channel transfer function via pilot tone [16]. However, the abovementione channel estimation schemes cannot be irectly applie to DHD ue to the changeable moe. One way to eal with this is to change channel estimation accoring to the selecte moe. Further research for this thir problem will help better implementation of DHD, but is not in the scope of this paper. The last problem is relate to the high complexity of DHD. Obviously DHD increases the system complexity, since it consiers three moes an each with ifferent resource allocation. To analyze the complexity theoretically, suppose that a frame has S subcarriers an T symbols with a suitable resource allocation algorithm for FDD. For simplicity, we assume a complexity measure as the number of times the algorithm is applie to fin optimal resource allocation. Then, the complexity of TDD becomes O(T), when an exhaustive search is consiere using the FDD algorithm as a sub-function. This is because the number of combinations to allocate symbols to DL an UL is T+1. Similarly, the complexity of DHD becomes O(T+2S). Since the complexity of DHD is epenent on the number of subcarriers in a frame, real-time implementation of DHD may be limite. However, note that the complexity increase is linear to the number of subcarriers. Thus, we expect that the evelopment of an efficient resource allocation algorithm for DHD together with a fast calculation evice will overcome the high complexity problem of DHD. 2.3 Avantages Avantages of DHD inclue flexibility, aaptability an efficiency. As explaine in previous sections, DHD can freely change the moe an the amount of resources for DL an UL. Thus, traffic asymmetry can be solve easily by the flexibility of DHD. DHD also provies optimize uplexing which is aapte to the given physical layer. Given the numbers of subcarriers an symbols in a frame an other parameters such as the size of the guar ban, DHD can aaptively support proper uplexing in the OFDMA system. The most important avantage of DHD is channel efficiency. DHD can efficiently 5

6 allocate subcarriers with high channel gain to proper DL an UL users. Figure 3 shows the efficiency of DHD over FDD an TDD. In the figure, otte an soli lines respectively represent the cases of subcarrier allocation to users with low an high channel gains. Clearly, the channel gain of each user is increase by using DHD. DHD applies rfdd moe in Figure 3(a) an FDD moe in Figure 3(b). When the channel gains of the four users are the same as in Figure 3(c), DHD/TDD moe presents higher channel performance than DHD/FDD. Obviously, the channel status of a real wireless environment may be more complex. However, even in the case, DHD can have more opportunities to increase system performance by flexibly allocating subcarriers with high channel gain to proper DL an UL users. 3. Resource Allocation in OFDMA/DHD In this section, we moel the OFDMA system. Then, resource allocation problem in DHD is consiere. 3.1 OFDMA System Moeling We consier a multi-user OFDMA system in a cell. Figure 2 shows the main functions of a BS in the system. In the figure, the system controller receives user s channel an traffic information as inputs, then allocates resources to DL an UL users. We are intereste in resource allocation to maximize system performance in OFDMA. The OFDMA system we consier has a frame structure which consists of S subcarriers an T symbols as shown in Figure 4. k own DL users an k up UL users are assume in the system an K = 1,2,..., k, k + 1,..., k + k } is the set of DL an UL { u user inices. For simplicity, we assume that a user can request either a DL service or a UL service in a frame. Each user has a rate requirement as in the set of rate requirements Θ = θ, θ,, θ, θ, θ }. { 1 2 k k + 1 k + ku In each moe, DHD allocates resources ifferently. Subcarriers are allocate to DL an UL in FDD (or rfdd) moe an symbols are allocate in TDD moe. To prevent interference between DL an UL, the guar ban is consiere with s g subcarriers for FDD (or rfdd) moe an the guar interval with t g symbols for TDD moe. Note that though the resources are ifferently allocate accoring to the moe, the resource allocation to DL an UL can be represente irrespective of the moe. To formalize this, we efine resource allocation vector as A = s, s, t, s, s, t }, { 1 2 u1 u2 u 6

7 s1 s 2 s u 1 where an are the first an the last inices of subcarriers allocate to DL, an s are those to UL, an t an t are the numbers of allocate symbols to DL u2 u an UL. The equations in Figure 4 are constraints the variables shoul satisfy in each moe. By assuming perfect channel estimation, the average channel gain for subcarrier i allocate to user, g, can be known. With the gain, the signal to noise ratio (SNR) of user with subcarrier i is obtaine as g p, where p is the allocate power to user with subcarrier i for one symbol perio. From the SNR, the achievable ata rate c of user with subcarrier i uring one symbol perio can be obtaine by the Shannon capacity formula as follows: ( 1 p ) c = log 2 + g. Note that the total available transmission power is limite in a BS (or a MS). To represent this, we efine Π = π, π, π,..., }, where π 0 an { π k u π are the total maximal transmission power of a BS an UL user + k, respectively. The above explaine terms are summarize in Table Formulation of Resource Allocation Problem In each frame of DHD, we nee to ecie the moe an the amount of resources to DL an UL. To eal with this, we formulate the resource allocation problem for DHD. The problem we consier is to maximize the total users ata rate with power an rate constraints an the restriction of allocating a subcarrier to one DL (or UL) user, consiering users channel conition, traffic information, an system parameters. Although our scope is restricte to the moe an the amount of resources to DL an UL, resource istribution to each user shoul be consiere to calculate the total users rate. Thus, we consier resource allocation for the moe an the amount of resources to DL an UL an user-level istribution for istribution to each user. First, for user-level istribution, we introuce the subcarrier assignment inicator, x, which is efine as 1, if subcarrier i is allocate to user x =, 0, otherwise 7

8 an assume the allocation vector A = s, s, t, s, s, t } is given. { 1 2 u1 u2 u Then, power constraints are expresse as in Equation (1) an (2). DL transmission is limite by total transmission power π 0 of a BS an UL transmission is limite by π of MS + k. s 2 k i= s 1 = 1 t p π 0 (1) s u 2 tu p π k for = k + 1, k + 2,, k i= su1 + k u (2) Another constraint in resource allocation is rate constraint. By using the achievable ata rate, c, which is efine in the previous section, the rate constraints for DL an UL are expresse as follows. s 2 t i= s 1 x log 1 ( + g p ) θ for = 1,2,, k (3) su 2 t u i= su1 x log 1 ( + g p ) θ for = k + 1, k + 2,, k + k (4) u Now, each DL (or UL) subcarrier has to be assigne to exactly one DL (or UL) user. This subcarrier assignment constraint can be expresse as k = 1 x 1 for i = s1, 2,s (5) k ku + x 1 for i = su1,,su 2 = k + 1 (6) x = 0 for i < s, i > s 2 an = 1,2,, k 1 (7) x = for i < s, i > s an = k + 1, k + 2,, k + k 0 u1 u 2 u (8) Recall that we have assume the resource allocation vector A is given to erive user-level istribution that satisfies the above constraints. Thus, we introuce a function, 8

9 Ω Π, Θ A ( ), which returns the set of all possible user-level istribution with A uner power constraint Π an rate requirement Θ. It can be expresse as follows. Ω Π ( A) =, Θ P {( X, ) (1) ~ (8)} (9) Here, X an P are the matrices for x an p, respectively. Finally, the total ata rate is efine with the following obective function f ( A, X, P) = su 2 k + k ( + g p ) + tux log( 1+ g p ) (10) s 2 k tx log 1 i= s 1 = 1 i= su1 = k + 1 an the formulation of DHD resource allocation problem (DRAP) results as follows. max A,X,P s.t f(a,x,p) ( X, P) Ω ( ) Π, Θ A (11) 4. Heuristic Algorithm for DRAP DRAP, which is a nonlinear problem, is har to solve even if the moe an the amount of resources allocate to DL an UL are fixe. Thus, we propose a heuristic algorithm which can solve DRAP effectively. We first suggest a simple greey heuristic to eal with user-level istribution, then provie an overall algorithm which use the user-level istribution heuristic as a sub-function. 4.1 User-Level Distribution In our algorithm, user-level istribution for DL an UL is necessary to calculate the total ata rate, when the moe an the amount of resources allocate to DL an UL are given. Recall that user-level istribution eals with how to allocate subcarriers an power to users uner power, rate an subcarrier assignment constraints. We assume users are accepte by an appropriate call amission control such that resources are sufficient to serve all accepte users. Our basic iea for the heuristic is to allocate a subcarrier to a user iteratively such that the rate requirement of each user is satisfie as much as possible. First, the caniate subcarrier is ecie for each user. For that purpose, we suggest a moifie water-filling proceure base on [9]. In the proceure, a temporary set is employe for each nonallocate subcarrier an water-filling is applie to the set to calculate the corresponing user rate. In the case of an UL user, the temporary set is mae by aing a non-allocate 9

10 subcarrier to the subcarriers alreay allocate to the user. Water-filling is applie with the maximal transmission power of the user. For a DL user, however, since all DL users share the available power of a BS, the temporary set is mae by aing a non-allocate subcarrier to all subcarriers allocate to DL users. The power for water-filling is the maximal transmission power of a BS. By selecting a subcarrier with the maximal rate among all non-allocate subcarriers, the caniate subcarrier of a user is obtaine. After the selection of the caniate subcarrier, we select a user to assign the subcarrier. If there exist users whose rate requirements are not satisfie, the user with the maximal gap between the requirement an the currently allocate rate is selecte. Otherwise, the user whose rate increase is maximize with the aitional subcarrier is selecte. The proceure is terminate when all subcarriers are assigne to users. Figure 5 shows the flow chart of the proceure. 4.2 Overall Algorithm Our algorithm is an iterative search proceure which selects the best moe an the best resource allocation vector to maximize the total ata rate of all users. For that purpose, the algorithm changes the amount of resources allocate to DL an UL in each moe an calculates the obective value with the heuristic in Section 4.1. By comparing the result of each moe, the best moe an the best resource allocation vector are obtaine. In each moe of the search, initial resource allocation vector A init is obtaine by allocating subcarriers in FDD (or rfdd) moe an symbols in TDD moe. To maximize the total ata rate that satisfies power constraints (1) an (2) in Section 3.2, we nee to balance the power efficiencies of DL an UL. Thus, in the initial solution, total subcarriers (or symbols) are allocate to DL an UL in proportion to the sum of SNR values of corresponing users. Note that the allocate power p shoul be calculate to obtain the SNR, which is efine in Section 3.1. To solve this problem, the average channel gain over all subcarriers an the equally istribute power are calculate, then the SNR value of each user is estimate by multiplying the two calculate values. The initial vector A init is sequentially upate to increase the total ata rate. On each upate, a subcarrier (or symbol) is remove from UL an ae to DL or vice versa. Two ifferent vectors which reflect the two cases are generate, then compare by the user-level istribution heuristic. The vector with a higher total rate is etermine. This hill-climbing upate proceure is continue until a better resource allocation vector cannot be foun. Figure 6 shows the pseuo-coe of FDD moe. In the pseuo-coe, 10

11 A init is obtaine by estimating the value of s with * given below. s FDD * = π 0 k u2 = 1 k + k u = k + 1 k k + k g π g + u s FDD π g = k + 1 g in the equation is the average channel gain of user over all subcarriers. In Figure 6, ( ) u Π, Θ A is a function of resource allocation vector A an returns the total rate obtaine by the user-level istribution heuristic. A represents the resource a b allocation vector which reflects that a variable a is b. Similarly to the pseuo-coe of FDD moe, those of rfdd an TDD moes are obtaine. In rfdd (or TDD), s u2 is change to s (or t ) with s * (or * ) given below. 2 s rfdd rfdd * = t TDD * = t TDD π 0 k k = 1 π 0 k g k = 1 k + k + g u = k + 1 π g 5. Simulation In this section, we perform simulations for various environments to compare DHD with other uplex schemes. We consier the OFDMA system in which a frame has 256 subcarriers an 20 symbols over 2GHz. The guar ban (or guar interval) is fixe to 10% of the total number of subcarriers (or symbols). In orer to consier frequency selective faing environment, we employ a 6-ray Rayleigh faing moel. The efault power of a BS an a MS are 1W an 200mW, respectively. The performance of DHD is evaluate by consiering five ifferent scenarios as shown in Table 2. Scenarios 1 an 2 consier ifferent power settings of a BS an MSs. Scenarios 3 an 4 eal with various rate requirements of DL an UL users. Scenario 5 is consiere to examine the effect of guar ban ratio, which is the ratio of the number of subcarriers for the guar ban to the total number of subcarriers. In the simulation, we compare DHD with TDD an FDD. The performance measure is the total ata rate by UL an DL users. 20 simulations are performe in each scenario 11

12 an the average is plotte in the figures. Figures 7 an 8 show results of scenarios 1 an 2, respectively. In each figure, DHD gives a better performance than TDD an FDD in all cases. This is because DHD efficiently selects the optimize moe an allocates the optimize resources accoring to the available power. In FDD, resource allocation cannot be aapte to the change of available power ue to its fixe banwith. TDD supports more efficient resource allocation than FDD. However, its inefficiency is ue to the constraint in which all subcarriers shoul be allocate to DL (or UL). The results of scenarios 3 an 4 are shown in Figures 9 an 10, respectively. Again, DHD outperforms other uplex schemes. It can be interprete that DHD supports efficient resource allocation irrespective of various rate requirements by changing its moe an allocating resources ynamically. Note that the total ata rate of each uplex scheme is slightly increasing with the lower rate requirement in Figure 9, which is in contrast with the result of Figure 10. The ecrease of rate requirement means the increase of opportunities to enhance system performance. Thus, the result illustrates that the user-level istribution is much more affecte by DL rate requirement than by UL rate requirement irrespective of the uplex metho. This is because DL users share the BS power an have more opportunities to increase the total rate than UL users. Figure 11, which is the result of scenario 5, shows that DHD can be aapte to various physical parameter settings. In the figure, the total rate of TDD is not changing, since TDD is affecte not by the guar ban but by the guar interval. On the other han, the total rate of FDD ecreases with the increase of the guar ban ratio. DHD is also affecte by the increase of the guar ban. However, even in the case, DHD emonstrates efficient resource allocation compare to other uplex schemes. Finally, the complexity an the total rate of the propose DHD are compare with other uplex schemes. The resource allocation for each uplex scheme is calculate by the user-level istribution algorithm suggeste in Section 4.1. The number of run times of the user-level istribution algorithm is consiere as the complexity measure. Exhaustive search which consiers all possible resource allocation combinations is also compare. Figure 12 shows the results of FDD, TDD, DHD an exhaustive search. As shown in the figure, the complexity of DHD is not serious consiering the throughput increase. The throughput gain of DHD over FDD an TDD well mitigates its complexity. However, the cost of the exhaustive search is too expensive to apply. From the results, we conclue that DHD with the propose algorithm enhances the system performance with a reasonable complexity. 12

13 6. Conclusion In this paper, a new uplex scheme, DHD, is propose, which changes the moe an the amount of allocate resources ynamically accoring to the traffic situation an user s channel status. System requirements to implement DHD are iscusse with the avantages of flexibility, aaptability an efficiency. Especially, efficiency is a noticeable avantage which can enhance system performance accoring to the given channel environment. To allocate resources in DHD, the OFDMA system is moele an DHD resource allocation problem (DRAP) is formulate which maximizes total ata rate with power, rate an subcarrier assignment constraints. An effective iterative search algorithm is suggeste to solve the DRAP. The algorithm initially allocates resource to DL an UL in proportion to the sum of SNR values, then searches for the best resource allocation vector by changing the amount of resources allocate to DL an UL. The propose uplex scheme, DHD, is simulate in various environments an compare with other uplex schemes. In the five ifferent scenarios, DHD always outperforms FDD an TDD. DHD efficiently increases its system performance in various settings by aapting its moe an allocating its resource ynamically. References 1. Sun JZ, Sauvola J, Howie D. Features in future: 4G visions from a technical perspective. IEEE GLOBECOM 2001; 6: Alexiou A, Avior D, Bosch P, Das S, Gupta P, Hochwal B, Klein TE, Ling J, Lozano A, Marzetta TL, Mukheree S, Mullener S, Papaias CB, Valenzuela RA, Viswanathan H. Duplexing, resource allocation an inter-cell coorination: esign recommenations for next generation wireless systems. Wireless Communications an Mobile Computing 2005; 5: Esmailzaeh R, Nakagawa M, Sourour EA. Time-ivision uplex CDMA communication. IEEE Personal Communications 1997; 4: Wong CY, Cheng RS, Letaief KB, Murch RD. Multiuser OFDM with aaptive subcarrier, bit an power allocation. IEEE Journal on Selecte Areas in Communications, October 1999; 17: Rhee W, Cioffi JM. Increase in capacity of multiuser OFDM system using ynamic subchannel allocation. In Proceeing of IEEE VTC 2000; 51(2): Pietrzyk S, Janssen GJM. Multiuser Subcarrier Allocation for QoS Provision in the OFDMA Systems. In Proceeing of IEEE VTC 2002; 56(2): Kim I, Lee HL, Kim B, Lee YH. On the use of linear programming for ynamic 13

14 subchannel an bit allocation in multinser OFDM. IEEE GLOBECOM 2001; 6; Kivanc D, Li G, Liu H. Computationally efficient banwith allocation an power control for OFDMA. IEEE Transactions on Wireless Communications, November 2003; 2(6): Jang J, Lee KB. Transmit Power Aaptation for Multiuser OFDM Systems. IEEE Journal on Selecte Areas in Communications, February 2003; 21(2): Li G, Liu H. Dynamic resource allocation with finite buffer constraint in broaban OFDMA networks. IEEE Wireless Communications an Networking, March 2003; 2: Kim K, Kim H, Han Y. Subcarrier an Power allocation in OFDMA Systems. In Proceeing of IEEE VTC 2004; 2: Wang W, Hwang KC, Lee KB, Bahk S. Resource allocation for heterogeneous service in multiuser OFDM systems. IEEE GLOBECOM 2004; 6: Han Z, Ji Z, Liu KJR. Low-complexity OFDMA channel allocation with Nash bargaining solution fairness. IEEE GLOBECOM 2004; 6: Munz G, Pfletschinger S, Speciel J. An efficient waterfilling algorithm for multiple access OFDM. IEEE GLOBECOM 2002, November 2002; 1: Song G, Li Y. Aaptive subcarrier an power allocation in OFDM base on maximizing utility. In Proceeing of IEEE VTC 2003-spring; Hanzo L, Munster M, Choi BJ, Keller T. OFDM an MC-CDMA for Broaban Multi-user Communications, WLANs an Broacasting, Wiley - IEEE Press, August

15 Figure 1. Three moes of DHD: FDD, reverse FDD, an TDD Figure 2. The base station block iagram Figure 3. Efficiency of DHD with higher channel gain 15

16 Figure 4. Frame structure escription for (a) FDD, (b) rfdd an (c) TDD moe S T s g the number of subcarriers the number of symbols uring one frame the number of subcarriers of which the guar ban consists t g the number of symbols of which the guar interval consists K the set of DL an UL user inices ( = 1,2,..., k, k + 1,..., k + k } ) Θ the set of require rates for DL an UL users ( } ) A g p c { u = { θ, θ2,, θk, θk + 1, θk + k 1 u the resource allocation vector ( = s, s, t, s, s, t }), where an su1 s u 2 { 1 2 u1 u2 u s1 s 2 ( an ) are the first an the last inices of subcarriers allocate to DL (UL), an t an t are the numbers of allocate symbols to DL an UL. u the average channel gain of user with subcarrier i the allocate power to user with subcarrier i for one symbol perio. Its matrix is P. the achievable ata rate of user with subcarrier i the set of total transmission power ( = { π 0, π1, π 2,..., π k }), where π u 0 an Π are the total maximal transmission power of a BS an UL user + k x the inicator variable of allocating subcarrier i to user. Its matrix is X. Table 1. Summary of the use terms π 16

17 Figure 5. The flow chart of user-level istribution heuristic Figure 6. The pseuo coe of search algorithm for FDD moe 17

18 Inex BS power MS power DL rate requirement UL rate requirement Guar ban ratio Scenario 1 0.8~1.2W 200mW % Scenario 2 1W 100~300mW % Scenario 3 1W 200mW 10~ % Scenario 4 1W 200mW 10 10~90 10% Scenario 5 1W 200mW ~14% Table 2. Five ifferent scenarios for the performance evaluation. Figure 7. The total rate for ifferent BS power Figure 8. The total rate for ifferent MS power 18

19 Figure 9. The total rate for ifferent DL rate requirement Figure 10. The total rate for ifferent UL rate requirement Figure 11. The total ata rate for ifferent guar ban ratio 19

20 Figure 12. The complexity an the total rate for FDD, TDD, DHD an exhaustive search 20