Fair scheduling in cellular systems in the presence of noncooperative mobiles

Size: px
Start display at page:

Download "Fair scheduling in cellular systems in the presence of noncooperative mobiles"

Transcription

1 1 Fair cheduling in cellular yte in the preence of noncooperative obile Veeraruna Kavitha +, Eitan Altan R. El-Azouzi + and Rajeh Sundarean Maetro group, INRIA, 2004 Route de Luciole, Sophia Antipoli, France + LIA, Univerity of Avignon, 339, chein de Meinajarie, Avignon, France IISc, Indian Intitute of Science, Bangalore, India Abtract We conider the proble of fair cheduling the reource to one of the any obile tation by a centrally controlled bae tation (BS. The BS i the only entity taking deciion in thi fraework baed on truthful inforation fro the obile on their radio channel. We tudy the well-known faily of paraetric -fair cheduling proble fro a gaetheoretic perpective in which oe of the obile ay be noncooperative. We firt how that if the BS i unaware of the noncooperative behavior fro the obile, the noncooperative obile becoe ucceful in natching the reource fro the other cooperative obile, reulting in unfair allocation. If the BS i aware of the noncooperative obile, a new gae arie with BS a an additional player. It can then do better by neglecting the ignal fro the noncooperative obile. The BS, however, becoe ucceful in eliciting the truthful ignal fro the obile only when it ue additional inforation (ignal tatitic. Thi new policy along with the truthful ignal fro obile for a Nah Equilibriu (NE which we call a Truth Revealing Equilibriu. Finally, we propoe new iterative algorith to ipleent fair cheduling policie that robutify the otherwie non-robut (in preence of noncooperation fair cheduling algorith. I. INTRODUCTION Short-ter fading arie in a obile wirele radio counication yte in the preence of catterer, reulting in tievarying channel gain. Variou cellular network have downlink hared data channel that ue cheduling echani to exploit the fluctuation of the radio condition (e.g. 3GPP HSDPA 2 and CDMA/HDR 8 or 1xEV-DO 1. A central cheduling proble in wirele counication i that of allocating reource to one of any obile tation that hare a coon radio channel. Much attention ha been given to the deign of efficient and fair cheduling chee that are centrally controlled by a bae tation (BS whoe deciion depend on the channel condition of each obile. Thee network ue variou fairne criteria (6, 4 called generalized -fair criteria to deign a cla of paraetric cheduling algorith (which we henceforth call a -fair cheduling algorith or -FSA. One pecial cae, proportional fair haring (PFS, ha been intenely analyzed a applied to the CDMA/HDR yte. See 12, 8, 7, 20, 3, 11, 17. Thee reult are applicable to the 3GPP HSDPA yte a well. Kuhner & Whiting 15 analyzed the PFS algorith uing tochatic approxiation technique and howed that the ayptotic averaged throughput can be driven to optiize a certain yte utility function (u of logarith of offet-rate. See alo Stolyar 21. The BS i the only entity taking deciion in all the above ethod, and the BS depend crucially on truthful reporting of their channel tate by the obile. For exaple, in the frequency-diviion duplex yte, the BS ha no direct inforation on the channel gain, but tranit downlink pilot, and relie on the obile reported value of gain on thee pilot for cheduling. A cooperative obile will truthfully report thi inforation to the BS. A noncooperative obile will however end a ignal that i likely to induce the cheduler to behave in a anner beneficial to the obile. In 13, 14 we analyzed efficient cheduling (a pecial cae with = 0, wherein the cheduler axiize the u throughput at the BS in preence of noncooperation uing a ignaling gae (22. The ignaling gae can be ued only for that pecial cae and an -fair cheduler with > 0 cannot be odeled by a ignaling gae: for -fair cheduler with > 0, the utilitie of the BS are not expected utilitie but are concave cobination of the uer expected utilitie. Further, -fair cheduler (with > 0 ha an inherent feedback in it tructure (ore detail in ection II and thi feedback ake the tudy difficult and different fro the above paper. Thi paper ha contribution to three ain area: Networking Apect: (1 We identify cae where noncooperation reult in an unfair bia in the channel aignent in favor of noncooperative obile, if the bae tation i unaware of the noncooperative behavior. (2 We characterize the liitation of the bae tation, and obtain condition under which even when it i aware of noncooperation, it i not able to hare fairly the reource. (3 We how that the ability to achieve fair haring, in the preence of noncooperation, depend on the paraeter. (4 We deign robut iterative algorith that, under uitable condition, fairly hare the reource even in the preence of noncooperative ignaling. Gae theoretical odeling: (1 We odel a noncooperative obile a a rational player that wihe to axiize it throughput. Since the -fair aignent i related to the axiization of a related utility function, one can view the BS a yet another player. We thu have a gae odel even if there i a ingle noncooperative obile. (2 We forulate three gae of which one i a concave gae. The forulation of the gae turn out to be urpriingly coplex. Except for the pecial cae of = 0 (where the gae can be hown to be equivalent to a atrix gae, the gae are defined over an infinite et of action. We are however able to prove the exitence and characterize the equilibriu policie

2 2 for two gae. (3 The third gae arie when the BS i unaware of noncooperation. BS only repond to the obile, but in a optial way. We could odel thi a a hierarchical gae where the obile are involved in a gae played at the higher level and the BS optiize oe utility at the lower level, unaware of the rationality of the obile. (4 To analyze iterative algorith, we conider a tochatic gae with ayptotic tie liit of the iterative algorith a cot criteria. Deign of the networking protocol baed on tochatic approxiation technique. (1 We analyze the paraetric -fair cheduling algorith (-FSA of 15 in preence of noncooperation. We identified it robutne propertie a a function of. (2 Uing the knowledge of channel and ignal tatitic, one can control the exce utilitie that the obile would have otherwie obtained by noncooperation. Thi i the baic idea behind robut policie. We then ue tochatic approxiation baed approach to cobine etiation (which replace the knowledge required and control to deign robut fair cheduling algorith. We firt otivate the proble uing a iple exaple. A Motivating exaple We conider two uer haring a coon channel. Uer 1 ha two channel tate with utilitie 7 and 3 occurring with probabilitie 0.33 and 0.67 repectively. Uer 2 ha contant channel with utility 4. The BS ha to aign the channel to one of the two uer for every realization of the channel tate and every uch aignent rule reult in a pair of uer average utilitie. The BS ue an -fair cheduler (decribed in the next ection to fair hare thee average utilitie. Firt we aue that both uer cooperate and report their individual channel tate correctly. In figure 1 and 2 (the cae with δ = 0 we plot the average utilitie obtained by uer under -fair cheduler a a function of the fairne paraeter. We ake the following obervation: (1 For every, the BS alway allocate the channel to uer 1 if he i in good tate. (2 For = 0, the expected hare of uer 1 ( i le than that of the uer 2 (( Thi correpond to efficient cheduling point. (3 For all value of, BS allocate the channel to uer 1 only when he i in good tate. (4 The expected hare of uer 1 increae while that of the uer 2 decreae a increae and eventually becoe equal. To achieve thi, the BS tart allocating the channel to the uer 1 even when that uer i in bad tate with increaing probability. The above cenario depend crucially on the truthful reporting of channel by the uer 1. Now, we conider the cenario when uer 1 i noncooperative and trie to increae hi utility. He declare to be in good tate 7 when actually in bad tate 3 with probability δ. BS now oberve the uer 1 to have good channel with better probability 0.33+δ 0.67 and will chedule a before but baed on reported channel condition. In figure 1, 2 we plot the reulting expected utilitie of both the uer a a function of fairne for δ = 0.1, δ = 0.5 repectively. We oberve that the utility of uer 1 for all value of i iproved in coparion with it cooperative utility. Thi alo reduce the utility of the uer 2 below it cooperative hare, utility of obile 1 δ = 0 utility of obile 2 δ = 0 utility of obile 1 δ = 0.1 utility of obile 2 δ = Fig. 1. Uer utilitie veru. for δ = utility of obile 1 δ = 0 utility of obile 2 δ = 0 utility of obile 1 δ = 0.5 utility of obile 2 δ = Fig. 2. Uer utilitie veru. for δ = 0.5 reulting in unfair allocation. Thi effect i een for all value of le than = 1.75, = 6.85 repectively for δ = 0.5, δ = 0.1. However, for alpha greater than the above value, uer 1 loe; in fact it utility get below it cooperative hare, while that of the uer 2 i uch above the later cooperative hare. The above exaple indicate the -fair cheduler: (1 ight be robut againt noncooperation for large value of (2 fail for aller value of. (3 the larger the δ the larger the aount of gain at = 0. (4 the larger the δ the aller the till which the obile gain. A increae, the two uer utilitie converge toward equal value at a rate that directly depend upon the difference at = 0. Thi i the reaon for the above obervation. An iportant point to note here i that, there i no threhold of beyond which the cheduler will be robut to all type of noncooperation, i.e., for all value of δ. However one can gue that for ax in fairne ( = the cheduler will be robut. The tudy of thi noncooperation and deign of robut policie will be the focu of our paper. II. THE PROBLEM SETTING AND -FAIR SCHEDULER The Downlink We conider the downlink of a wirele network with one bae tation (BS. There are M obile copeting for the downlink data channel. Tie i divided into all interval or lot. In each lot, one of the M obile i allocated the channel. Each obile can be in one of the tate h H, where H i finite valued. We aue fading characteritic to be independent acro the obile. Let h := h 1, h 2,, h M t be the vector of channel gain in a particular lot. The channel gain are ditributed according to: p h (h = M i=1 p h i (h i, where {p h ; M} repreent the tatitic of the obile channel. When the obile channel tate i h, it can achieve a axiu utility given by f(h. An exaple of utility i the rate f(h = r( = log(1 + h SNR where SNR capture the noinal received ignal-to-noie ratio under no channel variation. The deciion rule In every lot, the BS ha to ake cheduling deciion, i.e., allocate the downlink lot to one of the M uer, baed on the current realization of the channel tate vector h. For any et C, let P(C be the et of probability eaure on C. With that definition, a BS

3 3 deciion i a function β that aign to any given h an eleent in P({1, 2,, M}, the probability ditribution over the et of uer. Thu, β( h i the probability that the BS chedule current lot to obile given channel tate vector h. The -fairne criterion and cheduler We introduce the well known generalized -fair criterion (4 where the quantity that we wih to hare fairly i the expectation of the rando (intantaneou utilitie correponding to the aignent by the cheduler to the obile: G (β := M Γ (θ (β (1 =1 where θ (β := E h f(h β( h i the expected hare of obile under policy β and where the -fair function i { log(u, for = 1 Γ (u := u 1 1, for 1. One can view β(.., the cheduling policy, a a vector in R B pace, with B := M H, where H i the cardinality of the product pace H = Π M =1H. More preciely the doain of optiization i 1 : { } M D := β(.. : β( h = 1, β( h 0 for all h,. =1 The objective function G given by (1, i concave and continuou in β for each, while the doain D i copact and convex. Hence there alway exit a cooperative -fair cheduling BS trategy β : β (.. arg ax β D G (β. (2 Reark II-1: We ay view the BS chedule a a tatic optiization proble that correpond to a ingle choice of β. Notice that the optial chedule β axiize oe function of the expected hare of utilitie. Thi expected hare depend on aignent at all channel tate, and i therefore a joint optiization proble. Thi feature arie when > 0. When = 0 the proble i eparable, and the olution β ( h for a given h depend only on that h. Indeed, for > 0, the iplicit equation (3 below highlight a certain feedback that i abent in cae with = 0. Thi ake the preent tudy ignificantly different fro our previou work on efficient cheduling with trategic obile (13. Below we how a key (feedback property of fair cheduler. Define β a the vector (fixed point that atifie (if it exit the following: β ( h = 1 { arg ax j dγ (θ j(β f(h j} arg ax j dγ (θ j (β (3 f(h j where dγ (θ j (β := dγ du u=θj(β i the derivative of Γ with repect to (w.r.t. u, evaluated at θ j (β. We now have Lea 1: If there i a β atifying (3, then β i a global axiizer of the objective function in (2 over doain D and hence i an -fair cheduler. 1 For a given channel realization h the BS will chooe the obile randoly according to the eaure β(. h. Let Θ := θ 1 θ M T, Θ(β := θ 1 (β θ M (β T and Θ(D := {Θ(β : β D}. The ap Θ Γ (θ i trictly concave. Hence, there exit an unique axiizer (of the expected aigned hare over the convex et Θ(D: Θ = ax Θ Θ(D Γ (θ. (4 Hence, if there i a β atifying (3 then Θ = Θ(β. Further, any β which i a global axiu of the objective function (2 atifie the efficiency property: whenever f(h > f(h either β ( h, h > β ( h, h (5 or β ( h, h = β ( h, h {0, 1} for all h Π j H j and for all. Proof : Pleae refer to Appendix B. The above Lea 1 give the exact characterization of an optial olution of -fairne proble (3. It further talk about the efficiency of every poible -fair olution (5: the aignent for particular tate (h for any obile increae with the increae in the utility (f(h of the tate. Thi property i ued in the analyi under noncooperation. A part of Lea 1, regarding the poible olution (3, when retricted to proportional fairne, i already tated in 16. Reark II-2: The olution (3 explicitly how the feedback we entioned in Reark II-1. Thi olution ha already been ued in practical cenario (16 to achieve fair cheduling: The -fair olution for the dynaic etting with ergodic channel tate i the optial β that fair hare the tie average utilitie over a ingle realization of a whole aple path 2. In fact, the olution (3 under ergodicity can be ipleented by the following procedure: 1 At any tie lot k, obtain the cheduling deciion uing the current channel vector h k and uing the tie averaged aigned utilitie obtained till the lat tep {θ,k 1 } in place of {θ (β } of (3; 2 Update (in the obviou way the tie averaged aigned utilitie up to tep k {θ,k } uing the current cheduling deciion. III. PROBLEM FORMULATION UNDER NON COOPERATION In every lot, the BS need the knowledge of h for cheduling purpoe. In practice, obile etiate channel h uing the pilot ignal ent by BS. We aue perfect channel etiation. The obile end ignal { } to BS, a indication of the channel gain. Thu BS doe not have direct acce to channel tate h, but intead ha to rely on the obile for it. If the obile have elfih otive, they can ignal a better channel condition to grab the channel even when their channel i bad. The ain purpoe of thi paper i to tudy the effect of thi noncooperation on the -fair cheduler (2. We aue that 2 For ergodic channel under appropriate condition on the function g, 1 li K K K g(h k = E h g(h k=1 We are intereted in a particular function g(h = f(h β( h whoe average i exactly θ (β.

4 4 ignal are choen fro the channel pace itelf, i.e., H for all obile. We hall conider two type of cenario : Hierarchical gae G1: The BS i unaware of the poible noncooperative behavior fro the obile and applie the - fair cheduler (2 to the ignal = 1,, M t (a if they were the true channel value. The obile are aware of BS cheduler, ignal to optiize their own goal. When the bae tation i unaware, we odel thi gae a hierarchical gae with two level: where leader, the noncooperative obile, involve in a gae proble while BS, the follower doe the optiization. In thi gae, there i no coon knowledge: the bae tation doe not know the rationality of the obile. Thi gae i related to that dicued by Auann in 5 through any exaple. For everal year it ha been thought that the auption of coon knowledge of rationality for the player in the gae wa fundaental. It turn out that, in N- player gae, coon knowledge of rationality i not needed a an epiteic condition for equilibriu trategie (ee 5. A gae approach: The BS i odeled a an additional player in a one-hot gae. When the BS becoe aware of the poible noncooperation, it could ipleent better policie to do better. We firt conider a M player gae G2, where the BS chedule till only uing the ignal fro the obile. Becaue of it awarene, it could do better than the ituation of gae G1, but however will not be ucceful in copelling the obile to reveal the truthful ignal (Section V-A. In Section V-B we contruct ore intelligent (which require ore inforation BS policie which would be robut againt noncooperation: the new robut BS policie and the truthful ignal fro the obile for a Nah Equilibriu. Thi we refer a gae G3. We now introduce the iportant concept and definition that are ued in the paper. Thee are ore pecific to the firt two gae cenario. The correponding definition and concept ay vary lightly for the gae G3 and the difference are explained directly in Section V-B. Coon Knowledge : Channel tatitic {p h ; M} of all obile i a coon knowledge, i.e., known to all the obile and the BS. We alo aue that, the inforation about which obile are noncooperative, i a coon knowledge in cae of the lat two gae G2 and G3. If the rational BS doe not know which obile are cooperative, it will treat every obile a noncooperative. Mobile Policie : Soe obile (with indice 1 M 1 where (0 M 1 M are aued to be noncooperative. A policy of obile i a function {µ (. h } that ap a tate h to an eleent in P(H. BS Policie : A policy of the BS i a function which ap every ignal vector to a cheduler β P({1, 2,, M}. Thee policie are ued in ajor part of the paper, while ore coplicated policie are conidered in ection V-B. Utilitie for a given et of trategie : The intantaneou/aple utility of the obile depend only upon the true channel h and the BS deciion β and i given by : U (, h, β = 1 {β=} in{f(h, f( } 3. Define the following to exclude obile : h := h 1,, h 1, h +1,, h M, p h (h := Π j p hj (h j, µ ( h := Π j ;j M1 µ j ( j h j Π j ;j>m1 δ(h j = j. Alo define, µ = {µ ; M 1 } to repreent trategy profile: µ( h := Π 1 j M1 µ j ( j h j Π j>m1 δ(h j = j. With the above definition, each noncooperative uer chooe it trategy µ in uch a way a to axiize it own utility: U (µ, β = E h U (, h, β( µ( h Under the -fair criterion (1, the natural election of utility for BS will be: U BS(µ, β = (6 Γ (U (µ, β. (7 Throughout when arg ax S ha ore than one eleent, by i = arg ax S we ean i arg ax S. By j := arg ax S we ean that j i a choen eleent of arg ax S. ASA, ATA Utilitie : When obile ignal do not atch the true channel value, the gae under conideration will have two iportant average utilitie for any given trategy profile (µ, β : (1 average ignaled utilitie under aignent β (ASA utility, which a (ore intelligent BS can oberve, and (2 average true and aigned (ATA utility, which i the true average utility gained by the obile and whoe value cannot be etiated (a long a the obile i noncooperative at the BS. Thee are defined by U ASA (µ, β := E h U AT A (µ, β := E h f( µ( hβ( in{f(h, f( }µ( hβ( (8. (9 Indeed, it i eay to oberve that the utility of obile i it ATA utility, i.e., U(µ, β = U AT A (µ, β. Truth Revealing Strategy : In the following, by truth revealing trategy at obile we ean the trategy µ T ( h = 1 {=h } for all h, H, which ignal the true channel tate. Let µ T := (µ T 1,, µ T M. Under truthful trategie µ T, ATA and ASA utilitie coincide. For any BS policy β, if trategy profile (µ T, β for a Nah Equilibriu, then we call the NE a a Truth Revealing Equilibriu (TRE. 3 The obile achieve rate f(h even if the BS allocate it a higher rate f( becaue of the inflated ignal ent by the noncooperative obile. The jutification of thi i provided in detail in Appendix C (we ued iilar auption alo in 13, 14.

5 5 Cooperative Share : Bet repone of BS to truthful ignal µ T i any axiizer β of G (1. By Lea 1, the bet repone reult in unique axiu average ATA utilitie, θ c := θ (β = U (µ T, β, (10 which will be referred a Cooperative Share. Contrat between hierarchical optiization and the gae perpective: Recall that coputing a fair aignent by BS involve axiization of (1. Thu in the firt cenario, when obile chooe profile µ, the unaware BS fair hare ASA utilitie under µ by axiizing (12 (given in the next ection. However, what need fair haring i the ATA utilitie. Thi i achieved via the gae perpective, wherein the rational BS trie to fair hare the ATA utilitie gained by the obile. We tudy the variou cenario via three gae entioned above. IV. SCHEDULING UNDER NONCOOPERATION : HIERARCHICAL GAME PROBLEM G1 We conider the cenario in which the BS i unaware of the preence of noncooperative obile. A in cooperative etting, the BS chedule (uing optial cheduler (2 the channel to one of the obile uing the obile ignal, auing the to reflect the channel tate perfectly. The obile, aware of BS cheduler, axiize their utilitie. Utilitie of G1: For any given obile trategy profile µ, let the induced ignal probabilitie be repreented by p, i.e., p ( = h p h(hµ( h. Since the BS oberve p (intead of p h, it aue the expected hare of obile to be θ (µ, β := E p f( β( and hence axiize, UBS ASA (β, µ = Γ (θ (µ, β. (11 One can identify that θ (µ, β are the ASA utilitie. The equilibriu appropriate to thi cenario i Stackelberg Equilibriu. Stackelberg Equilibriu for G1: i a profile (β µ, µ which atifie the following for all : βµ = arg ax U ASA β µ = arg ax U AT A µ BS (β, µ, (12 ( (µ, µ, β (µ,µ.(13 We now preent oe exaple in which a uer deviate unilaterally fro µ T and increae it utility above it cooperative hare, reulting in unfair allocation. Thee exaple do not have TRE for G1, i.e., truthful trategy profile µ T i not a part of any Stackelberg Equilibriu of G1. In particular for (12, we conider -fair cheduler given by (3. Thi cheduler i a widely ued practical olution (ee Reark II-2, -FSA being one of the. A. Ayetric Exaple 1 Proportional fair cheduler ( = 1 : We continue with the otivating exaple given in Section I. Uer 1 ha a ingle tate with utility a. Uer 2 ha 2 tate with repective utilitie given by rb, b and with r > 1. The repective probabilitie to be in one of thee tate are p, (1 p with p (1/(1+r, 1/2. Uing (3, one can eaily etiate β, {θ (β } to be: β (2 a, rb = 1, β (1 a, b = 1, θ 1 (β = a(1 p and θ 2 (β = rbp. (14 Note that θ 1 (β, θ 2 (β are the obile cooperative hare. It i iportant to note here that β atifying (3 exit only if p (1/(1 + r, 1/2 a in thi cae : dγ (θ 2 (β rb = rb dγ (θ 2 (β b = b rbp > rbp < a a(1 p = dγ (θ 1 (β a a a(1 p = dγ (θ 1 (β a. Suppoe uer 2 ignal rb (when actually in tate b with probability q, i.e., µ 2 (rb b = q. Then uer axiu ASA rate (note that β q = β defined in (14 are: U ASA 1 (q, β q = (1 p qa, U AT A 2 (q, β q = rb(p + q repectively whenever rb (p + qrb > a a(1 p q > b rb(p + q. With thi, the obile 2 obtain an iproved ATA utility U2 AT A (q, βq = rbp+bq > θ 2 (β, i.e., obile 2 i ucceful in iproving it utility (above it cooperative hare by ignaling noncooperatively. The axiu poible value of q i q = (0.5 p. 2 Extenion to general : One can extend the above to general, an -fair cheduler atifying (3 exit if, (rb 1 p < a 1 (1 p < r(rb 1 p. Fro above, a increae, p for which (3 exit reduce and thu given (a, r, b, p, there exit a axiu ax, beyond which there doe not exit -fair cheduler of the type (3. However another type of alpha-fair cheduler exit; for exaple for ax-in fairne (when =, θ1 = θ2 a -fair cheduler {β (1 rb, a, β (1 b, a} given by: β a (1 rb, a = rbp + ap ; β (1 b, a = 0 if a(1 p < rbp β a(1 p rbp (1 b, a = (b + a(1 p ; β (1 rb, a = 0 ele. When -fair cheduler (3 exit the noncooperative obile benefit; the axiu q( atifie: (p + q( (rb 1 = a 1 (1 p q(. For exaple with a = 4, r = 3, b = 3, p = 0.33 the axiu for which -fair cheduler (3 exit i 7.9 and uer 1 can benefit by ignaling with q =.05 for all 4. 3 Generalization to ore tate and general : Conider two ayetric uer under the following auption : N.1 The cooperative -fair olution β (3 exit and without lo of generality let 1 = arg ax θ c. N.2 There exit an i > 1 uch that, η := inf dγ (θ1 c f(h 1,i 1 dγ (θ2 c f(h 2 > 0, h 2 H 2 where H 1 = {h 1,1,, h 1,N1 } are arranged uch that f(h 1,1 > f(h 1,2 > > f(h 1,N1.

6 6 Lea 2: Under auption N.1-N.2, there exit a non truth revealing policy µ δ 1 for obile 1 uch that it ATA utility U1 AT A (µ δ 1, (f, β i larger than it cooperative hare θ µ δ 1 c. 1 Proof : The proof i available in Appendix B. B. Syetric Cae We conider a iple yetric two obile exaple. The obile have two tate with utilitie a 1, a 2 occurring repectively with probabilitie p 1, p 2. Let a 1 = ra 2, p 1 = pp 2 with r > 1, p > 0. Under truthful ignaling, by Lea 1, an -fair optial BS policy (for any i given by: β (1 a 1, a 1 = 1/2 = β (1 a 2, a 2, β (1 a 1, a 2 = 1, β (1 a 2, a 1 = 0 with equal cooperative hare ( p θ 1 (β = θ 2 (β 2 = p 1p 2 a 1 + p 2 a ( p = p 2 2 r + 1 2a 2 + pr. 2 Without lo of generality ay obile 1 deviate unilaterally fro hi truthful trategy with µ 1 (a 1 a 2 = t. If obile 1 wa ucceful, hi reported rate would be greater than θ 1 (β and thi rate he would have obtained only when hi declared tate i a 1 with obile 2 being a 2. Thu, obile 1 will be ucceful with axiu ASA utilitie ( = 1: U1 ASA = (p 1 a 1 + p 2 ta 1 p 2 = (p + tp 2 2a 1 and U2 ASA = 1p 1 a 1 + p 2 (1 tp 2 a 2 = (pr + (1 tp 2 p 2 a 2 and the correponding ATA utility, U AT A 1 = (p 1 a 1 + p 2 ta 2 p 2 = (pr + tp 2 2a 2 if the following condition are et: a 1 U1 ASA > a 2 U2 ASA and θ 1 (β < U1 AT A, i.e., if t atifie: 1 1 > (p + tp 2 (pr + (1 tp 2 and p2 r + 1 < t. 2 C. Robutne at large For all value of, fair cheduler fail. However we ee a different phenoenon at higher. A increae to infinity, the fairne increae and the expected hare, i.e., ATA utilitie, of all the obile tend to becoing equal (18, provided all the obile ignal truthfully. However, in preence of noncooperation, it will be the ASA utilitie that tart becoing equal for higher value of. Thi reult in all the cooperative (ATA equal ASA utilitie obile getting equal ATA hare which will be bigger than that for the noncooperative (ATA are trictly le than ASA utilitie obile. Thu the -fair cheduler (2 itelf becoe ore and ore robut toward noncooperation a fairne factor increae, in pite of the BS unawarene of the noncooperation. 4 Thi effect i een 4 However a noticed in otivating exaple, we can only ay the ax in fairne will be robut againt all type of noncooperation fro obile and cannot identify an beyond which the cheduler will be robut. in the otivating exaple a well a in Figure 3 given in a later ection. In Figure 3, the noncooperative obile ATA utility diinihe a increae and goe below it cooperative hare beyond = 1.2 and further, the cooperative obile get ore hare than it cooperative hare for thee large value of. V. SCHEDULING UNDER NONCOOPERATION : GAME THEORETIC STUDY In thi ection the BS know about noncooperative behavior of obile and i conidered a an additional player reulting in the M player gae. A. BS Scheduling policie of ection IV : Gae G2 In contrat to ection IV, the BS know the obile that are noncooperative. The reulting gae i a one-hot concave gae: the utility of obile (6 i linear in it policy µ while that of the BS (7 i continuou and concave in it policy β. By 19, thi gae alway ha a NE 5 (µ, β which atifie, for all, µ = arg ax U µ ((µ, µ, β and β = arg ax U BS(µ, β. β For gae G2 we obtain a babbling equilibriu. We further how that G2 doe not have a TRE. 1 G2 ha Babbling NE : We will now how that thi gae ha a Nah equilibriu where the BS neglect the ignal fro the noncooperative uer. Let h >M1 := h M1+1, h M t repreent the channel tate of the cooperative obile. With θ >M1 (β := E h f(h β ( h >M1, the BS axiize: Γ ( θ >M1 (β. (15 We note here that for any non cooperative obile, θ >M1 (β = Ef(h E h >M 1 β ( h >M 1 for M1. A in Lea 1, there alway exit a β axiizing (15. Call one uch β by β >M1. Chooe any obile profile µ which atifie for all M 1, µ ( h = 0 for all h, with f( < f(h. It i eay to ee that (µ, β >M1 for a Nah Equilibriu. Note here that a noncooperative obile can obtain the utility θ >M1 (β >M1 only if it ignal better than it channel true value (a only in thi cae in{f(h, f( } = f(h and hence the requireent of above condition on the et of obile trategie. Thi i a NE at which the BS ignore the ignal fro the noncooperative obile and i iilar in ene to the Babbling equilibriu defined in the context of ignaling gae (22. Hence we choe to call thi alo a Babbling equilibriu. We end thi ubection with a ueful, practically ipleentable -fair cheduler involving only cooperative ignal 5 Note that when adding further concave contraint the gae reain concave even if the contraint are coupled 19. We thu obtain equilibriu alo for contrained verion of the gae. Exaple of uch contraint are: the (poible weighted u of throughput i bounded by a contant.

7 7 (if exit. Define (if exit, β >M1 ( h >M1 = A >M1 ( h >M1, β = arg ax j 1 { A >M 1(h >M 1,β >M 1 } A >M1 (h >M1, β >M1 dγ (θ >M1 j (βu j with u j = f(h j 1 {j>m1} + Ef(h j 1 {j M1}. Uing iilar tep which obtained (3, one can how that β >M1 i a axiizer of (15. 2 G2 ha No TRE : We now exaine the exitence of the deired TRE. The cae of = 0, the efficient cheduling i tudied in 13. In 13, G2 correponding to efficient cheduling wa odeled by a ignaling gae and it i hown that the gae G2 ha only babbling equilibriu a NE and hence doe not have a TRE. We will now conider the cae > 0. If the M player gae were to have a TRE, the correponding (equilibriu trategy of the BS, by definition the NE, hould be the bet repone to obile truthful trategie µ T and hence will be axiizer of UBS (µt, β = G (β. Hence, the bet repone for truth revealing trategy profile µ T indeed equal one of the axiizer of Lea 1, which atifie the efficiency property (5. Let β be any axiizer of Lea 1. The trategy profile (µ T, β doe not for a NE becaue: Let be any obile with non zero cooperative hare and let h be it channel value with larget utility, i.e., let h = arg ax h H f(h. The obile by changing it policy fro truthful ignal µ T to µ ( h := 1 for all h { = h}, increae it ATA utilitie a by (5 for any h and for any h h H, β ( h, h β ( h, h, and hence, U AT A ((µ T µ, β U AT A (µ T, β = ( p h (h β ( h, h f(h β ( hf(h h = ( p h (hf(h β ( h, h β ( h > 0. h Strict greater than zero reult in the lat line for all > 0, a all the obile obtain non zero utility under an alpha fair cheduler. Thu, the obile can iprove it utility by unilaterally oving away fro µ T, contradicting the definition of NE. Thu the BS, even when aware of the noncooperation, i not ucceful in eliciting the truthful ignal. In the following we contruct ore intelligent policie which induce a TRE. Hence, BS ha to ue ore intelligent cheduling algorith to be robut againt noncooperation. B. Robut BS Policie : Gae G3 ha TRE BS can etiate tatitic p after ufficient obervation of the obile ignal. We ue p to build robut policie for BS which give u the deired TRE. The policy of BS now ap every ordered pair of ignal and ignal tatitic (, p to an ordered pair (Φ, β = {(φ (, p, β(., p } with allocation φ ( f( for all. All the utilitie will change appropriately to include Φ, for exaple: U (µ, (Φ, β = E h in{φ (, f(h }µ( hβ( A profile (µ 1,, µ M 1, (Φ, β i a NE for G3 if, µ = arg ax U µ ((µ, µ, (Φ, β for all (Φ, β = arg ax U BS(µ, (Φ, β. (16 (Φ,β When BS know ignal tatitic, {p }, it can etiate the ASA utilitie for any cheduling policy and for any obile profile µ a: U ASA (µ, (Φ, β = U ASA (p, (Φ, β := E φ(β(. In the above the expectation i w.r.t. p. It can alo etiate their cooperative hare {θ c } of (10 uing it prior knowledge: the channel tatitic. We now propoe a robut policy at the BS which ue both thee average utilitie. The key idea i to deign a policy at BS which doe not allow the (average utility of any obile to be greater than θ c. When a noncooperative obile ue a ignaling trategy to iprove it ATA utility U AT A, even it ASA utility U ASA iprove. The BS can etiate U ASA of each of the obile and hence can ene the increae in the noncooperative obile ASA utility in coparion to it cooperative hare. The BS can enure none of the cooperative obile i allocated ore than it correponding cooperative hare: by allocating only a fraction and not the total ignaled utility at every aple. The fraction to be allocated, i et baed on the preent exce over the cooperative hare a follow: φ (, p, β := ( f( ( U ASA (p, (Φ, β θ c 1 {(f( (U ASA(p,(Φ,β θc >0} (17 for oe large value of. Hence, to enure that none of the obile get ore ASA utility than it cooperative hare, BS need to allocate (chooe Φ = {φ } to atify the following: U ASA (p, (Φ, β = E φ (, p, ββ(. (18 Both the equation (18 and (17 are atified if there exit a fixed point U ASA = U ASA (p, β which atifie: U ASA = E φ β( 1 {φ>0} ; (19 φ := f( ( U ASA θ c. With C f repreenting the upper bound on f, ( f( ( U ASA θ c 1{(f( (U ASA C f + θ c θc >0} for all and U ASA. Thu the ap U ASA φ β( 1 {φ>0} i bounded and continuou alot urely and hence by bounded convergence theore the ap of (19, U ASA E φ β( 1 {φ>0}, i continuou. Thu there exit an U ASA atifying the fixed point equation (19 by Brouwer fixed point theore 6. 6 Brouwer fixed point theore: Every continuou function f fro a cloed ball of a Euclidean pace to itelf ha a fixed point, i.e., an x which atifie x = f(x..

8 8 With the above allocation, ATA utility gained by obile, U AT A (µ, (Φ, β = E h, f gain (h,, p, ββ( (20 f gain (h,, p, β := in{f(h, φ (, p, β}. Equation (19 atifie : U ASA = Ef(β( 1+θc Eβ( 1 1+ E β( 1 (21 with 1 := 1 { U ASA <f(+ θc }. Hence, for any trategy profile (µ, β U ASA (p, (Φ, β θ c = E f( β( 1 { U ASA and hence, U AT A <f(+ θc } θ c 1 + E β( 1{ U ASA <f(+ θc C f o(1/, } (µ, (Φ, β U ASA (µ, (Φ, β θ c + o(1/. The above i true a, f gain (h,, µ, β φ (, µ, β. In other word, with new allocation (19 at BS, no obile can gain o(1/ ore than it cooperative hare for any pair (µ, β. Further, if BS ue any -fair cheduler β of (2, along with allocation policy (19, it i eay to check uing (21 and (20 that under truthful trategie (note p = p h U ASA (µ T, β1 = (µ T, β1 = θ c for all. Alo now, U AT A U ASA (p h, (Φ, β θ c = E h f(h β ( h1 E f(h β ( h 1 + E h β ( h1 = E h f(h β ( h1 { U ASA f(h+ θc } 1 + E h β. ( h1 The above indicate that U ASA (p h, (Φ, β θ c. If it wa trictly le than the cooperative hare, then the indicator in the nuerator of the econd line can never be true and hence there exit only one fixed point, θ c with (µ T, β. We have thu proved: Theore 1: If BS know cooperative hare {θ c } and the ignal tatitic {p }, the M 1 +1 player trategic gae ha an ɛ NE, i.e., TRE: ( µ T, ({φ (, p, β }, β (. In the coing ection, we will turn our attention to iterative algorith which can achieve a deired level of fairne even in the preence of oe noncooperative obile. We begin thi tak by firt tudying -FSA (15. VI. FAIR SCHEDULER ALGORITHM (-FSA Fro thi ection onward the channel tate h a well a the ignaled tate (the tate reported by the obile are continuou rando variable with tationary rate acro the tie, {r,k } k 1 = {f(h,k } k 1, {r,k } k 1 = {f(,k } k 1 for all, atifying the auption of Appendix A 7. Thi ection and the coing ection ue variou type of rate and hence the notation becoe coplicated. Thu a table (in table III of notation pecific to thee two ection i given in Appendix A, where all the rate notation are lited at one place. By auption A.3 of Appendix A, the rate are integrable and hence the ap Θ E h f(h 1 β(1 h, Θ,, E h f(h M β(m h, Θ, β( h, Θ = 1 {=arg ax j dγ (θ jf(h j} {arg ax j dγ (θ j f(h j } ha a fixed point Θ by Brouwer fixed point theore and β (. h := β(. h, Θ exactly atifie (3 and hence i a -fair olution. Thu with continuou rate we alway have fixed point -fair olution (3. We outlined an algorith to ipleent -fair cheduler (3 in Reark II-2 following Lea 1. The -FSA (15, a tochatic approxiation baed fair cheduling algorith, exactly follow thi outline (with Θ k θ := 1,k,, θ M,k, r k := r 1,k,, r M,k : θ,k = θ,k 1 + ɛ k I (r k, Θ k 1r,k θ,k 1 I(r, Θ = 1 {=arg axj dγ (d j+θ jr j} (22 = 1 {=arg axj r j(θ j+d j } where d are all poitive contant (added for tability. While aking deciion {I}, if there are ore than one uer attaining axiu, one of the axiizer i choen by the BS randoly. In 15, Th. 2.2, the author how that {θ,k } of (22, with 1, converge weakly to the unique liit point Θ that atifie E r I(r, Θ = θ for all. A cloe look at thi liit point (when we neglect {d } reveal that I(r, Θ i the -fair cheduler (3 and that Θ are the unique cooperative hare, {θ (β } = {θ c }. Thu, -FSA weakly converge to the unique point (cooperative hare that axiize the -fair criterion (1. A. Convergence of -FSA in preence of noncooperation The -FSA ue ignaled rate, r,k := f(,k and r k = r 1,k,, r M,k t to ake deciion, a in Section IV: θ,k = θ,k 1 + ɛ k I (r k, Θ k 1r,k θ,k 1. Thee ignaled rate reflect the tatitic p (intead of p h, there again i weak convergence, however thi tie to a different attractor correponding to p. It i very eay to ee a in Section IV that, when obile are noncooperative with 7 For undertanding the ayptotic liit of the dynaic algorith of thi ection we will need the reult correponding to the tatic etting of Section II. But, all the reult of Section II correpond to dicrete channel tate and rate. We aue that even for the ore general cae under tudy in thi ection, an -fair olution of the for (3 exit and that the correponding hare {θ c }are unique a in Lea 1. Sufficient condition for thi to occur are under tudy. Thi reult i required for howing that -FSA ayptotically converge to the cooperative hare (i.e., liit axiize the -fair criterion for all. In 15 Theore 2.3 doe thi job approxiately at leat for 1: any other aignent rule reult in a liit Θ with Γ (θ le than that correponding to cheduler {I } of -FSA (22. The iulation of thi ection alo confir the reult we obtained baed on thi auption.

9 9 profile µ, -FSA converge weakly to unique axiu ASA rate, { U ASA (µ, β µ } with β µ defined by (12. B. Failure of -FSA in preence of noncooperation A noted above, the -FSA (22 converge to the axiu ASA utility (under µ which need not be equal to the ATA utility, in the preence of noncooperation. However, to undertand the behavior of (22 in preence of noncooperation, one need to tudy the ayptotic true utilitie gained by the obile under (22. Toward thi, we conider a econd iteration running in parallel with (22, with only the intantaneou ignaled utility r,k replaced by the true intantaneou utility obtained by the obile, r,k := in{r,k, r,k }: θ,k = θ,k 1 + ɛ k I (r k, Θ k 1 r,k θ,k 1 (23 A in 15, one can how that θ,k converge weakly to U AT A (µ, βµ, the ATA utility under (µ, βµ. Thu, the ayptotic liit of -FSA equal axiu ASA utilitie of ection IV while the true utility adaptation (23 converge to the correponding ATA utilitie. Thee tie liit will thu have all the propertie of ection IV: the -FSA will fail for all and will be robut for large a dicued in ection IV. The only difference here i that the channel rate are continuou. C. Nuerical exaple Two ayetric uer are conidered in Figure 3. Let Z(σ 2 be a Rayleigh rando variable with denity f Z (z; σ 2 = ze z2 /2σ 2. Channel tate of uer 1 i conditional Rayleigh ditributed, i.e., h 1 f Z(z; 11 {z 2} dz. P (Z(1 2 Uer 2 ha alot a contant channel, h f Z(z; {z 2} dz. P (Z( The utilitie are the achievable rate f(h = log(1+h. Uer 1 i noncooperative with 1 (h = h(1 δ+2δ with δ = 0.9. We plot the liit of the -FSA, the liit of true utility adaptation (23 a function 8 of. We alo plot the cooperative hare, obtained by the liit of -FSA, i.e., with δ = 0. We oberve that the cooperative hare tend toward equal value a tend to infinity. Uer 1 i ucceful in gaining ore utility in coparion with it cooperative hare for all le than 1.2. Beyond 1.2, uer 1 actually loe and the lo increae a increae. The obervation are iilar to that in otivating exaple and indicate that -FSA i robut only for large. In table I, we conider a yetric exaple. In thi exaple, we conider the dicrete channel of ection IV. We could have contructed exaple with yetric continuou channel tate a in previou exaple and deontrate the failure of -FSA. But we note that the -FSA even with dicrete rate 8 The author in 15 analyze thee algorith only for 1. However we confir uing nuerou exaple that they work in fact for all value of, i.e., when all obile are cooperative the FSA for all converge to the unique hare which axiize the objective function ( FSA θ 1 c Mob 1 θ 2 c Mob 2 ASA Mob 1 (δ=0.9 ASA Mob 2 (δ=0.9 ATA Mob 1 (δ=0.9 ATA Mob 2 (δ= Fig. 3. -FSA : Maxiu ASA and correponding ATA hare veru Robut Fair SA θ 1 c Noncoop Mob θ 2 c Coop Mob ATA Noncoop Mob (δ = 0.9 ATA Coop Mob (δ = 0.9 ATA Noncoop Mob (δ = 0 ATA Coop Mob (δ = Fig. 4. Robut Policy: Maxial ASA and correponding ATA hare. work a explained in thi ection and hence thi exaple i given to deontrate the ae. We conider two uer, both of the having two channel tate with utilitie a1 = 4, a2 = 2 occurring with probabilitie p1 = 0.3, p 2 = 0.7 repectively. In thi exaple we work only with = 1, i.e., the proportional fair cheduler. Both uer have equal cooperative hare, θ 1 (β = θ 2 (β = Hence when both the obile report the channel tate truthfully, under FSA cheduler, the ayptotic throughput of both the obile converge to 1.51, i.e., li k θ,k = 1.51 for = 1, 2. Hence axiu proportionally fair BS (ayptotic utility i UBS = 2log(1.51 = µ 1 (a 1 a 2 True Rate t (U AT A 1, U AT A 2 U AT A log(u AT A 0 (Coop (1.51, (1.62, (1.72, (1.70, TABLE I A SYMMETRIC EXAMPLE IN WHICH FSA FAILS AGAINST NONCOOPERATION The uer 1 becoe noncooperative with µ 1 (a 1 a 2 = t. We ee that the uer 1 i ucceful in grabbing the channel ore often and increaing it utility in coparion with it cooperative hare. The ore he cheat (the ore t i the ore he gain (look at the ayptotic throughput U1 AT A, given in the econd colun in table I. He gain up to 12.5% ore than it cooperative hare. The cooperative uer, uer 2 ha lot due to the non cooperative obile reulting in unfair allocation. VII. ROBUST -FAIR ALGORITHMS : ROBUST FAIR SA We aw that -FSA fail in the preence of noncooperative uer. Hence, we propoe a robutification of -FSA againt noncooperation uing the policie of ubection V-B. In V-B, we propoed BS policie robut againt noncooperation and in thi ection we propoe tochatic approxiation baed algorith to converge toward the ASA utilitie of thoe policie given by (20. The policy of ection V-B require the knowledge of ignal tatitic p, which ha to be etiated. Baically the ethod decribed in thi ection (a i done by FSA cobine the etiation and control uing tochatic

10 10 approxiation baed ethod. We will how robutne of thee policie by uing appropriate gae theoretic tool. Robut Policy 1 : We now propoe a robutification of (22 againt noncooperation in the following : θ,k+1 = θ,k + ɛ k φ,k+1 I ( r k+1, Θ k θ,k φ,k+1 = ax { ( 0, r,k+1 ( θ,k θ c },(24 θ,0 = θ c (25 where the deciion I (r, Θ are ae a that in -FSA (22 and only the allocation Φ k := φ 1,k,, φ M,k T are ade robut. A in the cae of -FSA, to undertand the behavior of thi algorith we need the following iteration which etiate the true utilitie gained by the obile : ˆθ,k+1 = ˆθ,k + ɛ k ˆr,k+1I ( r k+1, Θ k ˆr,k+1 = in { r,k+1, φ },k+1 A. Analyi : ˆθ,k, (26 We analyze the robutne of the propoed algorith uing gae theoretical tool. Fix any. We conider a M 1 +1 player gae with utilitie defined by : U := li ˆθ,k+1 for all and U BS := Γ (U. k We analyze the liit of (26 uing ODE approxiation ethod (for e.g., 15, 9. A a firt tep, we obtain the following ODE approxiation reult. Theore 2: Aue that algorith (24, (26 atify auption A.1, A.2, and A.3 of Appendix A. For any initial condition, (Θ k, ˆΘ k converge weakly to the et of liit point of the olution of the ODE (for all M: θ = h (Θ θ, h (Θ = E φ I (r, Θ, (27 ˆθ = ĥ (Θ ˆθ, ĥ (Θ = E ˆr I (r, Θ. (28 Thee concluion hold whenever ɛ n 0, n ɛ n = and for oe n, li n up 0 l n ɛ n+l /ɛ n 1 = 0. Reark about the proof and the auption : Thi theore can be proved exactly in the ae way a i done for FSA by Theore 2.1 of Kuhner et al 15. The required auption A.1-3 are alo very iilar to that in 15 and thee are atified in alot the ae condition a entioned in 15. Hence, one can upper bound utilitie {U } by upper bounding all the attractor of ODE (28. Any attractor Θ of the ODE (27 atifie θ θ c = E ri (r, Θ I {φ >0} θ c 1 + E. Hence I (r, Θ I {φ >0} θ θ c C f E I (r, Θ I {φ >0} E = C f I (r, Θ I {φ >0} where C f i the upper bound on ignaled rate {r}. Thu, θ θ c + o(1/. Further, any attractor of ODE (28 atifie ˆθ = ĥ (Θ leading to ˆθ θ. Thu for any obile trategy profile µ, U w = ˆθ θ θ c + o(1/. (29 So, none of the uer, no atter what trategy they ue or the other ue, can gain ore than θ c. Under µ T, Θ c = θ1 c,, θm ct i the only zero of RHS of both the ODE (27, (28 a hown uing fixed point analyi in ection V-B (note here that I(r, Θ i the - fair cheduler β ( atifying (3. By virtue of Lea 4, one can eaily how that it will indeed be an attractor (for large enough value of for ODE (27 by howing that the derivative of { h (Θ θ ; M} i negative definite 9 near Θ c. And then it i iediate that Θ c i alo an attractor for the ODE (28. Thu, Θ c, i the only attractor of both the ODE under µ T. Thu U w = θ c for all under µ T. (30 Fro (29, (30, the robut policy (24 at BS together with the truth-revealing policy of uer for an ɛ-ne. Robut Policy 2 : The policie of previou ubection, Robut Policy 1 will not allow the ATA utility of any uer to go above the cooperative hare. Neverthele, when a uer i noncooperative, thee policie ay till reult in a lo for the cooperative uer: the noncooperative uer can till grab the channel fro other uer, without getting a gain becaue of the robut allocation policie (19. To avoid thi proble, we ay robutify the deciion a well: θ,k+1 = θ,k + ɛ k φ,k+1 I ( Φ k+1, Θ k θ,k. (31 The analyi of thi policy would be iilar to the policy 1. We need to change the auption of the Appendix A appropriately (need to replace the deciion I (r, Θ with I (Φ, Θ in all the place for ODE approxiation. However thee policie are ore coplicated and hence further analyi i ore difficult. We will till be able to go through all tep in the analyi exactly in a iilar way, except that we will not be able to how the uniquene of the attractor under truthful trategie µ T. However, we could enure the robutne of thee policie uing the nuerical exaple given below. The exaple alo how that thee policie outperfor Robut 9 Uing iilar tep a thoe ued for deriving (34 of Appendix B, one can eaily ee that ( h(θ θ = 1 E I θ (r, Θ c Θ=Θ c E (r 2 Π k j,i P r(a k (r, Θ c g j (κ(θc j + d j (θ c j + d+1 which i alway negative and whoe agnitude increae a increae for all while for any j ( h(θ θ = θ j Θ=Θ c E (r 2 Π k j,i P r(a k (r, Θc g j (κ(θc j + d j 1 (θ c + d whoe agnitude i bounded independent of. Define the ap H(Θ := h 1 (Θ,, h M (Θ T, the total derivative w.r.t Θ A := D Θ ( H(Θ Θ and the atrix B a a diagonal atrix coniting of only diagonal entire of atrix A. Note that atrix B ha all negative eigenvalue. Now, when 10, Corrollary III 2.6, pp. 63 i applied to atrice A, B (thi corollary copare the eigenvalue of the two atrice we get that atrix A becoe negative definite a the value of increae.

11 11 policy 1 in any way, while Robut policy 1 would be ipler to ipleent. 0.5 Robut Fair SA : Policy Robut Fair SA : Policy Nuerical exaple We continue with the exaple of Figure 3 (in which - FSA failed in Figure 4. We ue Robut Fair SA Policy 1 in place of -FSA. We et = We plot only the ATA utilitie for both value of δ = 0, δ = 0.9. We do not plot the ASA utilitie in thi figure a thee utilitie for all the cae are very cloe to cooperative hare Θ c. We ee that thi policy i indeed robut : 1 the tie liit of {θ,k } (which correpond to ASA utilitie are very cloe to the cooperative hare; 2 the tie liit of the ayptotic true (ATA utilitie are leer than the cooperative hare for the noncooperative obile. It i alo leer for cooperative obile, but the gap between the cooperative hare and the ATA utilitie i uch leer for a cooperative obile (plot correponding to δ = 0.9; 3 when all the obile are cooperative both the ASA a well a ATA utilitie are cloe to the cooperative hare for all the obile (plot correponding to δ = 0. In Figure 5, 6 we copare the two robut policie. Here h 1 f Z(z; 11 {z 2} dz P rob(z(1 2, h 2 f Z(z; 0.51 {z 2} dz P rob(z(0.5 2, f(h = log(1 + h and = Mobile 1, can be noncooperative uing 1 (h = h + (2 hδ with δ = 0.9. We ee fro the figure that both the policie are robut. Even with high value of δ = 0.9 (which indicate large aount of noncooperation both the policie do not allow the ATA utilitie to go beyond the cooperative hare. However the policy 2 i way better than the policy 1 a: 1 the noncooperative obile in fact i punihed by policy 2, it ATA utility i leer than the cooperative hare θ1 c (Figure 6, but the ae i not true for policy 1 (Figure 5. 2 the cooperative obile loe becaue of the noncooperation fro the other obile to uch greater extent in policy 1. Thi goe in line with the extra robutification built into the deciion aking by policy 2. When BS ue policy 1, the noncooperative obile i ucceful in grabbing the channel (alot alway with large value of δ = 0.9, however will not be able to gain uch fro it becaue of the robut allocation (19-(21. When obile i aware that he cant gain fro being noncooperative, he will a well have to tick to truthful ignaling, unle hi intention are that of jaing the other obile (in which cae the BS need to ue policy 2. However the policy 1 i eaier to ipleent than the policy 2 becaue of ipler deciion and ay alo have fater convergence. µ 1 (a 1 a 2 = True Rate U BS = t (U1 AT A, U2 AT A U AT A log(u AT A 0 (Coop (1.51, (1.30, (1.31, (1.32, TABLE II ROBUST POLICY 1 AGAINST NONCOOPERATION EXAMPLE OF TABLE I θ 1 c Mob 1 θ 2 c Mob 2 ASA Mob 1 (δ = 0.9 ASA Mob 2 (δ = 0.9 ATA Mob 1(δ = 0.9 ATA Mob 2(δ = 0.9 ASA Mob 1 (δ = 0 ASA Mob 2 (δ = 0 ATA Mob 1 (δ = 0 ATA Mob 2 (δ = Fig. 5. Robut Policy 1 : Maxial ASA and correponding ATA hare veru θ 1 c Mob 1 θ 2 c Mob 2 ASA Mob 1 (δ = 0.9 ASA Mob 2 (δ = 0.9 ATA Mob 1 (δ = 0.9 ATA Mob 1 (δ = 0.9 ASA Mob 1 (δ = 0 ASA Mob 2 (δ = 0 ATA Mob 1 (δ = 0 ATA Mob 2 (δ = Fig. 6. Robut Policy 2 : Maxial ASA and correponding ATA hare veru. In table II we continue with the yetric exaple of table I wherein the FSA fail. We ee once again that (even with dicrete and yetric condition the Robut Fair SA policy 1 i robut againt noncooperation, it doe not allow the noncooperative uer to iprove it ayptotic throughput. VIII. CONCLUSIONS We tudied centralized downlink traniion in a cellular network in the preence of noncooperative obile. Uing - fair cheduler, the BS ha to aign the lot to one of the any obile baed on truthful inforation fro obile about their tie-varying channel gain. A noncooperative obile ay irepreent it ignal to the BS o a to axiize hi throughput. We odeled a noncooperative obile a a rational player who wihe to axiize hi throughput. For thi gae, we identified everal cenario related to the awarene of BS. When the BS i unaware of thi noncooperative behavior, we odel thi gae a hierarchical gae with two level. We identify that, the preence of noncooperative uer, reult in an -fair bia in the channel aignent for all value of. A increae, an -fair cheduler becoe ore and ore robut to noncooperation irrepective of the awarene of BS and a ax-in fair cheduler i alway robut. When the BS i aware of the noncooperative obile, we characterized a babbling equilibriu which i obtained when both the BS and the noncooperative player ake no ue of the ignaling opportunitie. Thi gae ha no TRE (Truth Revealing Equilibriu. Uing additional knowledge of the tatitic of the ignal oberved at the BS, we built new robut policie to elicit the truthful ignal fro obile and achieve a Truth Revealing Equilibriu. We then tudied the popular, iterative fair cheduling algorith (which we called -FSA analyzed by Kuhner and Whiting in 15. We howed that - FSA fail under noncooperation. Finally, we propoed iterative robut fair haring to robutify the -FSA in the preence of noncooperation. Acknowledgent Thi project wa upported by the Indo-French Center for the Prootion of Advanced Reearch (IFCPAR, project IT-1 and by the INRIA aociation progra DAWN. The french co-author have alo been upported by the Bionet European project and the RNRT-ANR WiNEM project.

Fair scheduling in cellular systems in the presence of noncooperative mobiles

Fair scheduling in cellular systems in the presence of noncooperative mobiles Fair scheduling in cellular systes in the presence of noncooperative obiles Veeraruna Kavitha +, Eitan Altan R. El-Azouzi + and Rajesh Sundaresan Maestro group, INRIA, 004 Route des Lucioles, Sophia Antipolis,

More information

The Extended Balanced Truncation Algorithm

The Extended Balanced Truncation Algorithm International Journal of Coputing and Optiization Vol. 3, 2016, no. 1, 71-82 HIKARI Ltd, www.-hikari.co http://dx.doi.org/10.12988/ijco.2016.635 The Extended Balanced Truncation Algorith Cong Huu Nguyen

More information

Image Denoising Based on Non-Local Low-Rank Dictionary Learning

Image Denoising Based on Non-Local Low-Rank Dictionary Learning Advanced cience and Technology Letter Vol.11 (AT 16) pp.85-89 http://dx.doi.org/1.1457/atl.16. Iage Denoiing Baed on Non-Local Low-Rank Dictionary Learning Zhang Bo 1 1 Electronic and Inforation Engineering

More information

Scale Efficiency in DEA and DEA-R with Weight Restrictions

Scale Efficiency in DEA and DEA-R with Weight Restrictions Available online at http://ijdea.rbiau.ac.ir Int. J. Data Envelopent Analyi (ISSN 2345-458X) Vol.2, No.2, Year 2014 Article ID IJDEA-00226, 5 page Reearch Article International Journal of Data Envelopent

More information

Bayesian Reliability Estimation of Inverted Exponential Distribution under Progressive Type-II Censored Data

Bayesian Reliability Estimation of Inverted Exponential Distribution under Progressive Type-II Censored Data J. Stat. Appl. Pro. 3, No. 3, 317-333 (2014) 317 Journal of Statitic Application & Probability An International Journal http://dx.doi.org/10.12785/jap/030303 Bayeian Reliability Etiation of Inverted Exponential

More information

Ranking DEA Efficient Units with the Most Compromising Common Weights

Ranking DEA Efficient Units with the Most Compromising Common Weights The Sixth International Sypoiu on Operation Reearch and It Application ISORA 06 Xiniang, China, Augut 8 12, 2006 Copyright 2006 ORSC & APORC pp. 219 234 Ranking DEA Efficient Unit with the Mot Coproiing

More information

Assignment for Mathematics for Economists Fall 2016

Assignment for Mathematics for Economists Fall 2016 Due date: Mon. Nov. 1. Reading: CSZ, Ch. 5, Ch. 8.1 Aignment for Mathematic for Economit Fall 016 We now turn to finihing our coverage of concavity/convexity. There are two part: Jenen inequality for concave/convex

More information

On the Use of High-Order Moment Matching to Approximate the Generalized-K Distribution by a Gamma Distribution

On the Use of High-Order Moment Matching to Approximate the Generalized-K Distribution by a Gamma Distribution On the Ue of High-Order Moent Matching to Approxiate the Generalized- Ditribution by a Gaa Ditribution Saad Al-Ahadi Departent of Syte & Coputer Engineering Carleton Univerity Ottawa Canada aahadi@ce.carleton.ca

More information

Topic 7 Fuzzy expert systems: Fuzzy inference

Topic 7 Fuzzy expert systems: Fuzzy inference Topic 7 Fuzzy expert yte: Fuzzy inference adani fuzzy inference ugeno fuzzy inference Cae tudy uary Fuzzy inference The ot coonly ued fuzzy inference technique i the o-called adani ethod. In 975, Profeor

More information

AN EASY INTRODUCTION TO THE CIRCLE METHOD

AN EASY INTRODUCTION TO THE CIRCLE METHOD AN EASY INTRODUCTION TO THE CIRCLE METHOD EVAN WARNER Thi talk will try to ketch out oe of the ajor idea involved in the Hardy- Littlewood circle ethod in the context of Waring proble.. Setup Firt, let

More information

Research Article An Extension of Cross Redundancy of Interval Scale Outputs and Inputs in DEA

Research Article An Extension of Cross Redundancy of Interval Scale Outputs and Inputs in DEA Hindawi Publihing Corporation pplied Matheatic Volue 2013, rticle ID 658635, 7 page http://dx.doi.org/10.1155/2013/658635 Reearch rticle n Extenion of Cro Redundancy of Interval Scale Output and Input

More information

LEARNING DISCRIMINATIVE BASIS COEFFICIENTS FOR EIGENSPACE MLLR UNSUPERVISED ADAPTATION. Yajie Miao, Florian Metze, Alex Waibel

LEARNING DISCRIMINATIVE BASIS COEFFICIENTS FOR EIGENSPACE MLLR UNSUPERVISED ADAPTATION. Yajie Miao, Florian Metze, Alex Waibel LEARNING DISCRIMINATIVE BASIS COEFFICIENTS FOR EIGENSPACE MLLR UNSUPERVISED ADAPTATION Yajie Miao, Florian Metze, Alex Waibel Language Technologie Intitute, Carnegie Mellon Univerity, Pittburgh, PA, USA

More information

Investment decision for supply chain resilience based on Evolutionary Game theory

Investment decision for supply chain resilience based on Evolutionary Game theory Invetent deciion for upply chain reilience baed on Evolutionary Gae theory Xiaowei Ji(jixw@hut.edu.cn), Haijun Wang Manageent School Huazhong Univerity of Science and Technology Wuhan, Hubei, 4374, China

More information

Convergence of a Fixed-Point Minimum Error Entropy Algorithm

Convergence of a Fixed-Point Minimum Error Entropy Algorithm Entropy 05, 7, 5549-5560; doi:0.3390/e7085549 Article OPE ACCESS entropy ISS 099-4300 www.dpi.co/journal/entropy Convergence of a Fixed-Point Miniu Error Entropy Algorith Yu Zhang, Badong Chen, *, Xi Liu,

More information

CHAPTER 13 FILTERS AND TUNED AMPLIFIERS

CHAPTER 13 FILTERS AND TUNED AMPLIFIERS HAPTE FILTES AND TUNED AMPLIFIES hapter Outline. Filter Traniion, Type and Specification. The Filter Tranfer Function. Butterworth and hebyhev Filter. Firt Order and Second Order Filter Function.5 The

More information

Conservation of Energy

Conservation of Energy Add Iportant Conervation of Energy Page: 340 Note/Cue Here NGSS Standard: HS-PS3- Conervation of Energy MA Curriculu Fraework (006):.,.,.3 AP Phyic Learning Objective: 3.E.., 3.E.., 3.E..3, 3.E..4, 4.C..,

More information

Secretary problems with competing employers

Secretary problems with competing employers Secretary problem with competing employer Nicole Immorlica 1, Robert Kleinberg 2, and Mohammad Mahdian 1 1 Microoft Reearch, One Microoft Way, Redmond, WA. {nickle,mahdian}@microoft.com 2 UC Berkeley Computer

More information

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS 8.1 INTRODUCTION 8.2 REDUCED ORDER MODEL DESIGN FOR LINEAR DISCRETE-TIME CONTROL SYSTEMS 8.3

More information

Lecture 2 DATA ENVELOPMENT ANALYSIS - II

Lecture 2 DATA ENVELOPMENT ANALYSIS - II Lecture DATA ENVELOPMENT ANALYSIS - II Learning objective To eplain Data Envelopent Anali for ultiple input and ultiple output cae in the for of linear prograing .5 DEA: Multiple input, ultiple output

More information

Research Article An Adaptive Regulator for Space Teleoperation System in Task Space

Research Article An Adaptive Regulator for Space Teleoperation System in Task Space Abtract and Applied Analyi, Article ID 586, 7 page http://dx.doi.org/.55/24/586 Reearch Article An Adaptive Regulator for Space Teleoperation Syte in Tak Space Chao Ge, Weiwei Zhang, Hong Wang, and Xiaoyi

More information

THE BICYCLE RACE ALBERT SCHUELLER

THE BICYCLE RACE ALBERT SCHUELLER THE BICYCLE RACE ALBERT SCHUELLER. INTRODUCTION We will conider the ituation of a cyclit paing a refrehent tation in a bicycle race and the relative poition of the cyclit and her chaing upport car. The

More information

Preemptive scheduling on a small number of hierarchical machines

Preemptive scheduling on a small number of hierarchical machines Available online at www.ciencedirect.com Information and Computation 06 (008) 60 619 www.elevier.com/locate/ic Preemptive cheduling on a mall number of hierarchical machine György Dóa a, Leah Eptein b,

More information

Lecture 2 Phys 798S Spring 2016 Steven Anlage. The heart and soul of superconductivity is the Meissner Effect. This feature uniquely distinguishes

Lecture 2 Phys 798S Spring 2016 Steven Anlage. The heart and soul of superconductivity is the Meissner Effect. This feature uniquely distinguishes ecture Phy 798S Spring 6 Steven Anlage The heart and oul of uperconductivity i the Meiner Effect. Thi feature uniquely ditinguihe uperconductivity fro any other tate of atter. Here we dicu oe iple phenoenological

More information

Lecture 10 Filtering: Applied Concepts

Lecture 10 Filtering: Applied Concepts Lecture Filtering: Applied Concept In the previou two lecture, you have learned about finite-impule-repone (FIR) and infinite-impule-repone (IIR) filter. In thee lecture, we introduced the concept of filtering

More information

CONGESTION control is a key functionality in modern

CONGESTION control is a key functionality in modern IEEE TRANSACTIONS ON INFORMATION TEORY, VOL. X, NO. X, XXXXXXX 2008 On the Connection-Level Stability of Congetion-Controlled Communication Network Xiaojun Lin, Member, IEEE, Ne B. Shroff, Fellow, IEEE,

More information

An Exact Solution for the Deflection of a Clamped Rectangular Plate under Uniform Load

An Exact Solution for the Deflection of a Clamped Rectangular Plate under Uniform Load Applied Matheatical Science, Vol. 1, 007, no. 3, 19-137 An Exact Solution for the Deflection of a Claped Rectangular Plate under Unifor Load C.E. İrak and İ. Gerdeeli Itanbul Technical Univerity Faculty

More information

Section J8b: FET Low Frequency Response

Section J8b: FET Low Frequency Response ection J8b: FET ow Frequency epone In thi ection of our tudie, we re o to reiit the baic FET aplifier confiuration but with an additional twit The baic confiuration are the ae a we etiated ection J6 of

More information

The Features For Dark Matter And Dark Flow Found.

The Features For Dark Matter And Dark Flow Found. The Feature For Dark Matter And Dark Flow Found. Author: Dan Vier, Alere, the Netherland Date: January 04 Abtract. Fly-By- and GPS-atellite reveal an earth-dark atter-halo i affecting the orbit-velocitie

More information

Problem Set 8 Solutions

Problem Set 8 Solutions Deign and Analyi of Algorithm April 29, 2015 Maachuett Intitute of Technology 6.046J/18.410J Prof. Erik Demaine, Srini Devada, and Nancy Lynch Problem Set 8 Solution Problem Set 8 Solution Thi problem

More information

Evolutionary Algorithms Based Fixed Order Robust Controller Design and Robustness Performance Analysis

Evolutionary Algorithms Based Fixed Order Robust Controller Design and Robustness Performance Analysis Proceeding of 01 4th International Conference on Machine Learning and Computing IPCSIT vol. 5 (01) (01) IACSIT Pre, Singapore Evolutionary Algorithm Baed Fixed Order Robut Controller Deign and Robutne

More information

TP A.30 The effect of cue tip offset, cue weight, and cue speed on cue ball speed and spin

TP A.30 The effect of cue tip offset, cue weight, and cue speed on cue ball speed and spin technical proof TP A.30 The effect of cue tip offet, cue weight, and cue peed on cue all peed and pin technical proof upporting: The Illutrated Principle of Pool and Billiard http://illiard.colotate.edu

More information

Investigation of application of extractive distillation method in chloroform manufacture

Investigation of application of extractive distillation method in chloroform manufacture Invetigation of application of etractive ditillation ethod in chlorofor anufacture Proceeding of uropean Congre of Cheical ngineering (CC-6) Copenhagen, 16-20 Septeber 2007 Invetigation of application

More information

Lecture 17: Frequency Response of Amplifiers

Lecture 17: Frequency Response of Amplifiers ecture 7: Frequency epone of Aplifier Gu-Yeon Wei Diiion of Engineering and Applied Science Harard Unierity guyeon@eec.harard.edu Wei Oeriew eading S&S: Chapter 7 Ski ection ince otly decribed uing BJT

More information

Exponentially Convergent Controllers for n-dimensional. Nonholonomic Systems in Power Form. Jihao Luo and Panagiotis Tsiotras

Exponentially Convergent Controllers for n-dimensional. Nonholonomic Systems in Power Form. Jihao Luo and Panagiotis Tsiotras 997 Aerican Control Conference Albuquerque, NM, June 4-6, 997 Exponentially Convergent Controller for n-dienional Nonholonoic Syte in Power For Jihao Luo and Panagioti Tiotra Departent of Mechanical, Aeropace

More information

Bogoliubov Transformation in Classical Mechanics

Bogoliubov Transformation in Classical Mechanics Bogoliubov Tranformation in Claical Mechanic Canonical Tranformation Suppoe we have a et of complex canonical variable, {a j }, and would like to conider another et of variable, {b }, b b ({a j }). How

More information

A New Model and Calculation of Available Transfer Capability With Wind Generation *

A New Model and Calculation of Available Transfer Capability With Wind Generation * The Eighth International Sypoiu on Operation Reearch and It Application (ISORA 09) Zhangjiajie, China, Septeber 0, 009 Copyright 009 ORSC & APORC, pp. 70 79 A New Model and Calculation of Available Tranfer

More information

ADAPTIVE CONTROL DESIGN FOR A SYNCHRONOUS GENERATOR

ADAPTIVE CONTROL DESIGN FOR A SYNCHRONOUS GENERATOR ADAPTIVE CONTROL DESIGN FOR A SYNCHRONOUS GENERATOR SAEED ABAZARI MOHSEN HEIDARI NAVID REZA ABJADI Key word: Adaptive control Lyapunov tability Tranient tability Mechanical power. The operating point of

More information

Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions

Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions Stochatic Optimization with Inequality Contraint Uing Simultaneou Perturbation and Penalty Function I-Jeng Wang* and Jame C. Spall** The John Hopkin Univerity Applied Phyic Laboratory 11100 John Hopkin

More information

On the Stability Region of Congestion Control

On the Stability Region of Congestion Control On the Stability Region of Congetion Control Xiaojun Lin and Ne B. Shroff School of Electrical and Computer Engineering Purdue Univerity, Wet Lafayette, IN 47906 {linx,hroff}@ecn.purdue.edu Abtract It

More information

Control Systems Analysis and Design by the Root-Locus Method

Control Systems Analysis and Design by the Root-Locus Method 6 Control Sytem Analyi and Deign by the Root-Locu Method 6 1 INTRODUCTION The baic characteritic of the tranient repone of a cloed-loop ytem i cloely related to the location of the cloed-loop pole. If

More information

The Impact of Imperfect Scheduling on Cross-Layer Rate. Control in Multihop Wireless Networks

The Impact of Imperfect Scheduling on Cross-Layer Rate. Control in Multihop Wireless Networks The mpact of mperfect Scheduling on Cro-Layer Rate Control in Multihop Wirele Network Xiaojun Lin and Ne B. Shroff Center for Wirele Sytem and Application (CWSA) School of Electrical and Computer Engineering,

More information

Clustering Methods without Given Number of Clusters

Clustering Methods without Given Number of Clusters Clutering Method without Given Number of Cluter Peng Xu, Fei Liu Introduction A we now, mean method i a very effective algorithm of clutering. It mot powerful feature i the calability and implicity. However,

More information

24P 2, where W (measuring tape weight per meter) = 0.32 N m

24P 2, where W (measuring tape weight per meter) = 0.32 N m Ue of a 1W Laer to Verify the Speed of Light David M Verillion PHYS 375 North Carolina Agricultural and Technical State Univerity February 3, 2018 Abtract The lab wa et up to verify the accepted value

More information

Social Studies 201 Notes for November 14, 2003

Social Studies 201 Notes for November 14, 2003 1 Social Studie 201 Note for November 14, 2003 Etimation of a mean, mall ample ize Section 8.4, p. 501. When a reearcher ha only a mall ample ize available, the central limit theorem doe not apply to the

More information

Alternate Dispersion Measures in Replicated Factorial Experiments

Alternate Dispersion Measures in Replicated Factorial Experiments Alternate Diperion Meaure in Replicated Factorial Experiment Neal A. Mackertich The Raytheon Company, Sudbury MA 02421 Jame C. Benneyan Northeatern Univerity, Boton MA 02115 Peter D. Krau The Raytheon

More information

Performance Analysis of Sub-Rating for Handoff Calls in HCN

Performance Analysis of Sub-Rating for Handoff Calls in HCN I. J. Counication, Network and Syte Science, 29, 1, 1-89 Publihed Online February 29 in SciRe (http://www.scirp.org/journal/ijcn/). Perforance Analyi of Sub-Rating for Handoff Call in HCN Xiaolong WU 1,

More information

2FSK-LFM Compound Signal Parameter Estimation Based on Joint FRFT-ML Method

2FSK-LFM Compound Signal Parameter Estimation Based on Joint FRFT-ML Method International Conerence on et eaureent Coputational ethod (C 5 FS-F Copound Signal Paraeter Etiation Baed on Joint FF- ethod Zhaoyang Qiu Bin ang School o Electronic Engineering Univerity o Electronic

More information

A New Predictive Approach for Bilateral Teleoperation With Applications to Drive-by-Wire Systems

A New Predictive Approach for Bilateral Teleoperation With Applications to Drive-by-Wire Systems 1 A New Predictive Approach for Bilateral Teleoperation With Application to Drive-by-Wire Syte Ya-Jun Pan, Carlo Canuda-de-Wit and Olivier Senae Departent of Mechanical Engineering, Dalhouie Univerity

More information

A Constraint Propagation Algorithm for Determining the Stability Margin. The paper addresses the stability margin assessment for linear systems

A Constraint Propagation Algorithm for Determining the Stability Margin. The paper addresses the stability margin assessment for linear systems A Contraint Propagation Algorithm for Determining the Stability Margin of Linear Parameter Circuit and Sytem Lubomir Kolev and Simona Filipova-Petrakieva Abtract The paper addree the tability margin aement

More information

1-D SEDIMENT NUMERICAL MODEL AND ITS APPLICATION. Weimin Wu 1 and Guolu Yang 2

1-D SEDIMENT NUMERICAL MODEL AND ITS APPLICATION. Weimin Wu 1 and Guolu Yang 2 U-CHINA WORKHOP ON ADVANCED COMPUTATIONAL MODELLING IN HYDROCIENCE & ENGINEERING epteber 9-, Oxford, Miiippi, UA -D EDIMENT NUMERICAL MODEL AND IT APPLICATION Weiin Wu and Guolu Yang ABTRACT A one dienional

More information

EXTENDED STABILITY MARGINS ON CONTROLLER DESIGN FOR NONLINEAR INPUT DELAY SYSTEMS. Otto J. Roesch, Hubert Roth, Asif Iqbal

EXTENDED STABILITY MARGINS ON CONTROLLER DESIGN FOR NONLINEAR INPUT DELAY SYSTEMS. Otto J. Roesch, Hubert Roth, Asif Iqbal EXTENDED STABILITY MARGINS ON CONTROLLER DESIGN FOR NONLINEAR INPUT DELAY SYSTEMS Otto J. Roech, Hubert Roth, Aif Iqbal Intitute of Automatic Control Engineering Univerity Siegen, Germany {otto.roech,

More information

PPP AND UNIT ROOTS: LEARNING ACROSS REGIMES

PPP AND UNIT ROOTS: LEARNING ACROSS REGIMES PPP AND UNIT ROOTS: LEARNING ACROSS REGIMES GERALD P. DYWER, MARK FISHER, THOMAS J. FLAVIN, AND JAMES R. LOTHIAN Preliinary and incoplete Abtract. Taking a Bayeian approach, we focu on the inforation content

More information

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281 72 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 28 and i 2 Show how Euler formula (page 33) can then be ued to deduce the reult a ( a) 2 b 2 {e at co bt} {e at in bt} b ( a) 2 b 2 5 Under what condition

More information

AP CHEM WKST KEY: Atomic Structure Unit Review p. 1

AP CHEM WKST KEY: Atomic Structure Unit Review p. 1 AP CHEM WKST KEY: Atoic Structure Unit Review p. 1 1) a) ΔE = 2.178 x 10 18 J 1 2 nf 1 n 2i = 2.178 x 10 18 1 1 J 2 2 6 2 = 4.840 x 10 19 J b) E = λ hc λ = E hc = (6.626 x 10 34 J )(2.9979 x 10 4.840 x

More information

Social Studies 201 Notes for March 18, 2005

Social Studies 201 Notes for March 18, 2005 1 Social Studie 201 Note for March 18, 2005 Etimation of a mean, mall ample ize Section 8.4, p. 501. When a reearcher ha only a mall ample ize available, the central limit theorem doe not apply to the

More information

IN SUPERVISING the correct operation of dynamic plants,

IN SUPERVISING the correct operation of dynamic plants, 1158 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 14, NO. 6, NOVEMBER 2006 Nonlinear Fault Detection and Iolation in a Three-Tank Heating Syte Raffaella Mattone and Aleandro De Luca, Senior Meber,

More information

3.185 Problem Set 6. Radiation, Intro to Fluid Flow. Solutions

3.185 Problem Set 6. Radiation, Intro to Fluid Flow. Solutions 3.85 Proble Set 6 Radiation, Intro to Fluid Flow Solution. Radiation in Zirconia Phyical Vapor Depoition (5 (a To calculate thi viewfactor, we ll let S be the liquid zicronia dic and S the inner urface

More information

Improving Efficiency Scores of Inefficient Units. with Restricted Primary Resources

Improving Efficiency Scores of Inefficient Units. with Restricted Primary Resources Applied Matheatical Science, Vol. 3, 2009, no. 52, 2595-2602 Iproving Efficienc Score of Inefficient Unit with Retricted Priar Reource Farhad Hoeinzadeh Lotfi * Departent of Matheatic, Science and Reearch

More information

LOAD AND RESISTANCE FACTOR DESIGN APPROACH FOR FATIGUE OF MARINE STRUCTURES

LOAD AND RESISTANCE FACTOR DESIGN APPROACH FOR FATIGUE OF MARINE STRUCTURES 8 th ACE pecialty Conference on Probabilitic Mechanic and tructural Reliability PMC2000-169 LOAD AND REITANCE FACTOR DEIGN APPROACH FOR FATIGUE OF MARINE TRUCTURE Abtract I.A. Aakkaf, G. ACE, and B.M.

More information

Chapter Landscape of an Optimization Problem. Local Search. Coping With NP-Hardness. Gradient Descent: Vertex Cover

Chapter Landscape of an Optimization Problem. Local Search. Coping With NP-Hardness. Gradient Descent: Vertex Cover Coping With NP-Hardne Chapter 12 Local Search Q Suppoe I need to olve an NP-hard problem What hould I do? A Theory ay you're unlikely to find poly-time algorithm Mut acrifice one of three deired feature

More information

(b) Is the game below solvable by iterated strict dominance? Does it have a unique Nash equilibrium?

(b) Is the game below solvable by iterated strict dominance? Does it have a unique Nash equilibrium? 14.1 Final Exam Anwer all quetion. You have 3 hour in which to complete the exam. 1. (60 Minute 40 Point) Anwer each of the following ubquetion briefly. Pleae how your calculation and provide rough explanation

More information

A Simplified Methodology for the Synthesis of Adaptive Flight Control Systems

A Simplified Methodology for the Synthesis of Adaptive Flight Control Systems A Simplified Methodology for the Synthei of Adaptive Flight Control Sytem J.ROUSHANIAN, F.NADJAFI Department of Mechanical Engineering KNT Univerity of Technology 3Mirdamad St. Tehran IRAN Abtract- A implified

More information

A First Digit Theorem for Square-Free Integer Powers

A First Digit Theorem for Square-Free Integer Powers Pure Matheatical Science, Vol. 3, 014, no. 3, 19-139 HIKARI Ltd, www.-hikari.co http://dx.doi.org/10.1988/p.014.4615 A Firt Digit Theore or Square-Free Integer Power Werner Hürliann Feldtrae 145, CH-8004

More information

SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD

SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD S.P. Teeuwen, I. Erlich U. Bachmann Univerity of Duiburg, Germany Department of Electrical Power Sytem

More information

Linear Quadratic Stochastic Differential Games under Asymmetric Value of Information

Linear Quadratic Stochastic Differential Games under Asymmetric Value of Information Preprint of the 2th World Congre The International Federation of Automatic Control Touloue, France, July 9-4, 27 Linear Quadratic Stochatic Differential Game under Aymmetric Value of Information Dipankar

More information

Control of industrial robots. Decentralized control

Control of industrial robots. Decentralized control Control of indutrial robot Decentralized control Prof Paolo Rocco (paolorocco@poliiit) Politecnico di Milano Dipartiento di Elettronica, Inforazione e Bioingegneria Introduction Once the deired otion of

More information

Relevance Estimation of Cooperative Awareness Messages in VANETs

Relevance Estimation of Cooperative Awareness Messages in VANETs Relevance Etiation of Cooperative Awarene Meage in VANET Jakob Breu Reearch and Developent Dailer AG Böblingen, Gerany Eail: jakobbreu@dailerco Michael Menth Departent of Coputer Science Univerity of Tübingen

More information

SIMM Method Based on Acceleration Extraction for Nonlinear Maneuvering Target Tracking

SIMM Method Based on Acceleration Extraction for Nonlinear Maneuvering Target Tracking Journal of Electrical Engineering & Technology Vol. 7, o. 2, pp. 255~263, 202 255 http://dx.doi.org/0.5370/jeet.202.7.2.255 SIMM Method Baed on Acceleration Extraction for onlinear Maneuvering Target Tracking

More information

White Rose Research Online URL for this paper: Version: Accepted Version

White Rose Research Online URL for this paper:   Version: Accepted Version Thi i a repoitory copy of Identification of nonlinear ytem with non-peritent excitation uing an iterative forward orthogonal leat quare regreion algorithm. White Roe Reearch Online URL for thi paper: http://eprint.whiteroe.ac.uk/107314/

More information

Mobile Communications TCS 455

Mobile Communications TCS 455 Mobile Counication TCS 455 Dr. Prapun Sukopong prapun@iit.tu.ac.th Lecture 24 1 Office Hour: BKD 3601-7 Tueday 14:00-16:00 Thurday 9:30-11:30 Announceent Read Chapter 9: 9.1 9.5 Section 1.2 fro [Bahai,

More information

MODE SHAPE EXPANSION FROM DATA-BASED SYSTEM IDENTIFICATION PROCEDURES

MODE SHAPE EXPANSION FROM DATA-BASED SYSTEM IDENTIFICATION PROCEDURES Mecánica Coputacional Vol XXV, pp. 1593-1602 Alberto Cardona, Norberto Nigro, Victorio Sonzogni, Mario Storti. (Ed.) Santa Fe, Argentina, Noviebre 2006 MODE SHAPE EXPANSION FROM DATA-BASED SYSTEM IDENTIFICATION

More information

EFFECT ON PERSISTENCE OF INTRA-SPECIFIC COMPETITION IN COMPETITION MODELS

EFFECT ON PERSISTENCE OF INTRA-SPECIFIC COMPETITION IN COMPETITION MODELS Electronic Journal of Differential Equation, Vol. 2007(2007, No. 25, pp. 0. ISSN: 072-669. URL: http://ejde.math.txtate.edu or http://ejde.math.unt.edu ftp ejde.math.txtate.edu (login: ftp EFFECT ON PERSISTENCE

More information

Beta Burr XII OR Five Parameter Beta Lomax Distribution: Remarks and Characterizations

Beta Burr XII OR Five Parameter Beta Lomax Distribution: Remarks and Characterizations Marquette Univerity e-publication@marquette Mathematic, Statitic and Computer Science Faculty Reearch and Publication Mathematic, Statitic and Computer Science, Department of 6-1-2014 Beta Burr XII OR

More information

Root Locus Diagram. Root loci: The portion of root locus when k assume positive values: that is 0

Root Locus Diagram. Root loci: The portion of root locus when k assume positive values: that is 0 Objective Root Locu Diagram Upon completion of thi chapter you will be able to: Plot the Root Locu for a given Tranfer Function by varying gain of the ytem, Analye the tability of the ytem from the root

More information

Optimal revenue management in two class pre-emptive delay dependent Markovian queues

Optimal revenue management in two class pre-emptive delay dependent Markovian queues Optimal revenue management in two cla pre-emptive delay dependent Markovian queue Manu K. Gupta, N. Hemachandra and J. Venkatewaran Indutrial Engineering and Operation Reearch, IIT Bombay March 15, 2015

More information

HIGH-THROUGHPUT DUAL-MODE SINGLE/DOUBLE BINARY MAP PROCESSOR DESIGN FOR WIRELESS WAN

HIGH-THROUGHPUT DUAL-MODE SINGLE/DOUBLE BINARY MAP PROCESSOR DESIGN FOR WIRELESS WAN HIGH-THROUGHPUT DUAL-MODE SINGLE/DOUBLE BINARY MAP PROCESSOR DESIGN FOR WIRELESS WAN Chun-Yu Chen Cheng-Hung Lin and An-Yeu (Andy) Wu Graduate Intitute of Electronic Engineering and Departent of Electrical

More information

A BATCH-ARRIVAL QUEUE WITH MULTIPLE SERVERS AND FUZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH

A BATCH-ARRIVAL QUEUE WITH MULTIPLE SERVERS AND FUZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH Mathematical and Computational Application Vol. 11 No. pp. 181-191 006. Aociation for Scientific Reearch A BATCH-ARRIVA QEE WITH MTIPE SERVERS AND FZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH Jau-Chuan

More information

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004 18.997 Topic in Combinatorial Optimization April 29th, 2004 Lecture 21 Lecturer: Michel X. Goeman Scribe: Mohammad Mahdian 1 The Lovaz plitting-off lemma Lovaz plitting-off lemma tate the following. Theorem

More information

Theoretical Computer Science. Optimal algorithms for online scheduling with bounded rearrangement at the end

Theoretical Computer Science. Optimal algorithms for online scheduling with bounded rearrangement at the end Theoretical Computer Science 4 (0) 669 678 Content lit available at SciVere ScienceDirect Theoretical Computer Science journal homepage: www.elevier.com/locate/tc Optimal algorithm for online cheduling

More information

New bounds for Morse clusters

New bounds for Morse clusters New bound for More cluter Tamá Vinkó Advanced Concept Team, European Space Agency, ESTEC Keplerlaan 1, 2201 AZ Noordwijk, The Netherland Tama.Vinko@ea.int and Arnold Neumaier Fakultät für Mathematik, Univerität

More information

Predicting the Performance of Teams of Bounded Rational Decision-makers Using a Markov Chain Model

Predicting the Performance of Teams of Bounded Rational Decision-makers Using a Markov Chain Model The InTITuTe for ytem reearch Ir TechnIcal report 2013-14 Predicting the Performance of Team of Bounded Rational Deciion-maer Uing a Marov Chain Model Jeffrey Herrmann Ir develop, applie and teache advanced

More information

Codes Correcting Two Deletions

Codes Correcting Two Deletions 1 Code Correcting Two Deletion Ryan Gabry and Frederic Sala Spawar Sytem Center Univerity of California, Lo Angele ryan.gabry@navy.mil fredala@ucla.edu Abtract In thi work, we invetigate the problem of

More information

Adaptive Radar Signal Detection with Integrated Learning and Knowledge Exploitation

Adaptive Radar Signal Detection with Integrated Learning and Knowledge Exploitation Integrated Learning and Knowledge Exploitation Hongbin Li Departent of Electrical and Coputer Engineering Steven Intitute of Technology, Hoboken, NJ 73 USA hli@teven.edu Muralidhar Rangaway AFRL/RYAP Bldg

More information

Copyright 1967, by the author(s). All rights reserved.

Copyright 1967, by the author(s). All rights reserved. Copyright 1967, by the author(). All right reerved. Permiion to make digital or hard copie of all or part of thi work for peronal or claroom ue i granted without fee provided that copie are not made or

More information

ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION. Xiaoqun Wang

ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION. Xiaoqun Wang Proceeding of the 2008 Winter Simulation Conference S. J. Maon, R. R. Hill, L. Mönch, O. Roe, T. Jefferon, J. W. Fowler ed. ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION Xiaoqun Wang

More information

DIFFERENTIAL EQUATIONS

DIFFERENTIAL EQUATIONS Matheatic Reviion Guide Introduction to Differential Equation Page of Author: Mark Kudlowki MK HOME TUITION Matheatic Reviion Guide Level: A-Level Year DIFFERENTIAL EQUATIONS Verion : Date: 3-4-3 Matheatic

More information

4 Conservation of Momentum

4 Conservation of Momentum hapter 4 oneration of oentu 4 oneration of oentu A coon itake inoling coneration of oentu crop up in the cae of totally inelatic colliion of two object, the kind of colliion in which the two colliding

More information

Frequency Response Analysis of Linear Active Disturbance Rejection Control

Frequency Response Analysis of Linear Active Disturbance Rejection Control Senor & Tranducer, Vol. 57, Iue, October 3, pp. 346-354 Senor & Tranducer 3 by IFSA http://www.enorportal.co Freuency Repone Analyi of Linear Active Diturbance Reection Control Congzhi HUANG, Qing ZHENG

More information

Rigorous analysis of diffraction gratings of arbitrary profiles using virtual photonic crystals

Rigorous analysis of diffraction gratings of arbitrary profiles using virtual photonic crystals 2192 J. Opt. Soc. A. A/ Vol. 23, No. 9/ Septeber 2006 W. Jiang and R. T. Chen Rigorou analyi of diffraction grating of arbitrary profile uing virtual photonic crytal Wei Jiang and Ray T. Chen Microelectronic

More information

Lecture 4 Topic 3: General linear models (GLMs), the fundamentals of the analysis of variance (ANOVA), and completely randomized designs (CRDs)

Lecture 4 Topic 3: General linear models (GLMs), the fundamentals of the analysis of variance (ANOVA), and completely randomized designs (CRDs) Lecture 4 Topic 3: General linear model (GLM), the fundamental of the analyi of variance (ANOVA), and completely randomized deign (CRD) The general linear model One population: An obervation i explained

More information

LOW ORDER MIMO CONTROLLER DESIGN FOR AN ENGINE DISTURBANCE REJECTION PROBLEM. P.Dickinson, A.T.Shenton

LOW ORDER MIMO CONTROLLER DESIGN FOR AN ENGINE DISTURBANCE REJECTION PROBLEM. P.Dickinson, A.T.Shenton LOW ORDER MIMO CONTROLLER DESIGN FOR AN ENGINE DISTURBANCE REJECTION PROBLEM P.Dickinon, A.T.Shenton Department of Engineering, The Univerity of Liverpool, Liverpool L69 3GH, UK Abtract: Thi paper compare

More information

15 N 5 N. Chapter 4 Forces and Newton s Laws of Motion. The net force on an object is the vector sum of all forces acting on that object.

15 N 5 N. Chapter 4 Forces and Newton s Laws of Motion. The net force on an object is the vector sum of all forces acting on that object. Chapter 4 orce and ewton Law of Motion Goal for Chapter 4 to undertand what i force to tudy and apply ewton irt Law to tudy and apply the concept of a and acceleration a coponent of ewton Second Law to

More information

Optimal Coordination of Samples in Business Surveys

Optimal Coordination of Samples in Business Surveys Paper preented at the ICES-III, June 8-, 007, Montreal, Quebec, Canada Optimal Coordination of Sample in Buine Survey enka Mach, Ioana Şchiopu-Kratina, Philip T Rei, Jean-Marc Fillion Statitic Canada New

More information

Nonlinear Single-Particle Dynamics in High Energy Accelerators

Nonlinear Single-Particle Dynamics in High Energy Accelerators Nonlinear Single-Particle Dynamic in High Energy Accelerator Part 6: Canonical Perturbation Theory Nonlinear Single-Particle Dynamic in High Energy Accelerator Thi coure conit of eight lecture: 1. Introduction

More information

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION Vol. VIII Decoupling Control - M. Fikar

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION Vol. VIII Decoupling Control - M. Fikar DECOUPLING CONTROL M. Fikar Department of Proce Control, Faculty of Chemical and Food Technology, Slovak Univerity of Technology in Bratilava, Radlinkého 9, SK-812 37 Bratilava, Slovakia Keyword: Decoupling:

More information

ANALOG REALIZATIONS OF FRACTIONAL-ORDER INTEGRATORS/DIFFERENTIATORS A Comparison

ANALOG REALIZATIONS OF FRACTIONAL-ORDER INTEGRATORS/DIFFERENTIATORS A Comparison AALOG REALIZATIOS OF FRACTIOAL-ORDER ITEGRATORS/DIFFERETIATORS A Coparion Guido DEESD, Technical Univerity of Bari, Via de Gaperi, nc, I-7, Taranto, Italy gaione@poliba.it Keyword: Abtract: on-integer-order

More information

Convex Hulls of Curves Sam Burton

Convex Hulls of Curves Sam Burton Convex Hull of Curve Sam Burton 1 Introduction Thi paper will primarily be concerned with determining the face of convex hull of curve of the form C = {(t, t a, t b ) t [ 1, 1]}, a < b N in R 3. We hall

More information

Reliability Analysis of Embedded System with Different Modes of Failure Emphasizing Reboot Delay

Reliability Analysis of Embedded System with Different Modes of Failure Emphasizing Reboot Delay International Journal of Applied Science and Engineering 3., 4: 449-47 Reliability Analyi of Embedded Sytem with Different Mode of Failure Emphaizing Reboot Delay Deepak Kumar* and S. B. Singh Department

More information

Chapter 2 Sampling and Quantization. In order to investigate sampling and quantization, the difference between analog

Chapter 2 Sampling and Quantization. In order to investigate sampling and quantization, the difference between analog Chapter Sampling and Quantization.1 Analog and Digital Signal In order to invetigate ampling and quantization, the difference between analog and digital ignal mut be undertood. Analog ignal conit of continuou

More information

Memoryle Strategie in Concurrent Game with Reachability Objective Λ Krihnendu Chatterjee y Luca de Alfaro x Thoma A. Henzinger y;z y EECS, Univerity o

Memoryle Strategie in Concurrent Game with Reachability Objective Λ Krihnendu Chatterjee y Luca de Alfaro x Thoma A. Henzinger y;z y EECS, Univerity o Memoryle Strategie in Concurrent Game with Reachability Objective Krihnendu Chatterjee, Luca de Alfaro and Thoma A. Henzinger Report No. UCB/CSD-5-1406 Augut 2005 Computer Science Diviion (EECS) Univerity

More information