On the Performance of Randomized Power Control Algorithms in Multiple Access Wireless Networks Arash Behzad, Izhak Rubin†, and Julan Hsu Electrical Engineering Department University of California (UCLA) Los Angeles, CA 90095-1594 {abehzad, rubin}@ee.ucla.edu

Abstract—In the ground radio environment, the propagation loss and multipath fading can make the received power of packets at an intended destination differ by an order of magnitude. Such a condition allows the designer to take advantage of the capture effect, under which the packet with the strongest power may be successfully received even in the presence of other interfering packets. The latter phenomena is reinforced under Randomized Power Control Algorithms: every node, based on the power density function f(P), independently selects a power level (from a discrete or continuous set of available power levels) for transmission of each one of its packets.

I.

In wireless communications, random access mechanisms are commonly used for applications where there is a rapid delivery of small number of messages arriving at irregular intervals. Random access mechanisms are indispensable approaches in wireless networks due to their ease of implementation and low overhead. Specifically, such mechanisms are able to provide a low packet delay as long as the traffic load remains moderate. In a common type of application a large number of mobile users intend to transmit their messages to a single central station. Simple to analyze and straightforward to implement, Aloha is a widely studied and deployed medium access control protocol: almost all deployed cellular systems use Aloha (or one of its variants) as a mean to request network access for mobile users ([3], [4]). Similarly, Aloha is widely used for reservation requests in two-way messaging systems [6].

In this paper, we develop and investigate a novel power control algorithm (the Fair Randomized Power Control algorithm; FRPC) to enhance the throughput level of the slotted Aloha medium access control mechanism. FRPC is developed based on the analysis of the capture effect for which the transmitter captures the channel only if the observed signal-to-interference and noise ratio (SINR) at receiver is above a certain threshold. We first analyze the sensitivity of the generic randomized power control algorithm with respect to expected value and standard deviation of the power density function. In particular, based on the one-tailed version of Chebyshev’s inequality, we derive a closed-form expression for an upper bound on the probability of capture (and an upper bound on the aggregate throughput) in the network. We then apply the latter result to the process of designing the FRPC algorithm for the slotted Aloha MAC mechanism. Our simulation results indicate that the FRPC algorithm leads to significant throughput gain while providing fairness for different mobile users in the system.

Aloha was first introduced by Abramson [1], where it was determined that the maximum channel utilization under “pure Aloha” is about 18 percent (1/2e) of the channel bandwidth. The slotted variation was then introduced by Roberts, showing its capacity is doubled (over unslotted Aloha), to about 36 percent (1/e) [2]. In classical slotted Aloha it is assumed that a transmission is successful if and only if no other user attempts to use the channel during the same slot ([2], [3]). This is a reasonable assumption when all the packets are received under nearly equal power levels. Clearly, in the ground radio environment such an assumption is somewhat pessimistic as the propagation loss (for example, due to differences in the length of the transmission path) and multipath fading can make the received power of packets at the central station differ by an order of magnitude. These physically-induced power variations can enhance the capture effect, under which the packet with the strongest power may be successfully received even in the presence of other interfering packets [12]. The latter phenomena have been observed to actually improve the throughput of slotted Aloha, yet it leads to a high degree of unfairness in the network ([13], [12], [9]). In turn, one can consider the use of deliberate user-induced randomness in the transmit power level in an effort to improve the throughput of the slotted Aloha mechanism. We refer to such an approach as

1

Keywords: Medium Access Control (MAC); power control; slotted Aloha; wireless networks.



This work was supported by Office of Naval Research (ONR) under Contract No. N00014-01-C-0016, as part of the AINS (Autonomous Intelligent Networked Systems) project, by the National Science Foundation (NSF) under Grant No. ANI-0087148, and by University of California/Conexant MICRO Grant No. 04-100.

IEEE Communications Society / WCNC 2005

INTRODUCTION

707

0-7803-8966-2/05/$20.00 © 2005 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on March 11, 2009 at 04:07 from IEEE Xplore. Restrictions apply.

load). Conventionally, a newly arrived packet is transmitted in the first slot after its arrival; packets which are not received successfully are buffered and retransmitted after a random delay. However, similar to many studies in the literature, we assume that the newly arrived packets and packets awaiting transmissions are treated identically. That is, in every slot, each node transmits a packet according to a Bernoulli process with parameter q, 0 < q ≤ 1, whether it is a new packet or retransmitted packet. The set of available transmit power levels (for each node) is subject to the underlying circuitry design. Such a set might include a continuous spectrum of power levels (e.g. the Prism chipset) or only a handful of discrete power levels (e.g. the Cisco/Aironet cards). Every node independently selects a power level (from a discrete or continuous set of available power levels) for transmission of each one of its packets ([7]-[12]).

the Randomized Power Control Algorithms ([7]-[12]): every node, based on the power density function f(P), independently selects a power level (from a discrete or continuous set of available power levels) for transmission of each one of its packets. In this paper, we develop and investigate a novel power control algorithm (the Fair Randomized Power Control algorithm; FRPC) to enhance the throughput level of the slotted Aloha medium access control mechanism. FRPC is developed based on the analysis of the capture effect for which the transmitter captures the channel only if the observed signal-to-interference and noise ratio (SINR) at receiver is above a certain threshold. We first analyze the sensitivity of the generic randomized power control algorithm with respect to expected value and standard deviation of the power density function. In particular, based on the one-tailed version of Chebyshev’s inequality, we derive a closed-form expression for an upper bound on the probability of capture (and an upper bound on the aggregate throughput) in the network. We then apply the latter result to the process of designing the FRPC algorithm for the slotted Aloha MAC mechanism. Our simulation results indicate that the FRPC algorithm leads to significant throughput gain. Furthermore, FRPC provides fairness for different mobile users in the system so that the maximum difference between the throughput levels attained by the user nodes reduces.

Without loss of generality, throughout this paper we assume that if there are M simultaneous transmissions in a slot (from mobiles to the central station), then the set of simultaneous transmitters is equal to {1,2, " , M } . A transmission from node k to the central station is received successfully if the signal-tointerference and noise ratio (SINR) at the receiver is not less than the minimum required threshold γ , i.e. Gk Pk M

N+

(1)

i i

i =1 i≠k

The rest of the paper is organized as follows. In section II, we present the system model and the underlying assumptions. The mathematical analysis is elaborated in section III. Numerical results and conclusions are presented in section IV and section V, respectively.

where M is the number of simultaneous transmissions in the underlying slot, and N denotes the thermal noise power at the receiver. This model is generically referred to as the SINRbased Interference Model ([5]).

II. SYSTEM MODEL

III.

There are n mobile nodes in the network, labeled 1, …, n, which strive to communicate with a central station. All nodes operate in the same channel and transmit under a fixed data rate R. Nodes are equipped with identical half-duplex radios and omni-directional antennas. Let Gi denote the channel gain (incorporating distance dependent power attenuation and effects of link loss phenomena such as fading and shadowing) for transmission from node i to the central station. The value of Gi can be dynamically estimated by the mobile user in a usual manner through transmission of a test packet (containing the associated transmit power level) form a mobile to the central station [15]. The channel access protocol is slotted Aloha. That is, the time axis is divided into identical synchronized time slots whose duration is assumed to be equal to the transmission time of a packet (which is assumed to be fixed) plus some overhead duration that includes the maximum propagation delay. Each user can transmit its packet only at the beginning of a slot. It is assumed that all nodes always have packets waiting for being transmitted (i.e., heavy traffic

IEEE Communications Society / WCNC 2005

∑G P

≥ γ, k = 1,2,..., M ,

MATHEMATICAL ANALYSIS

A general randomized power control algorithm with a power density function f(P) is considered. Momentarily assume that the channel gains for all the n users are identical (G).2 Assuming that the minimum power level with a positive power density function is sufficiently large that results in a signal-tonoise ratio level larger than γ , the throughput (TH) of the network (in packets per slot) can be calculated as the following, TH = Pr{exactly one transmission in a slot} + Pr{capture in a slot} = nq(1 − q)

n −1

+ Pr{capture in a slot}.

(2) (3)

Since the probability of transmission (q) is independent of the power distribution, the latter does not have any effect on the first term of equation (3). Consequently, the only impact of the power distribution on the throughput is crystallized in 2

708

This assumption is later relaxed in the paper.

0-7803-8966-2/05/$20.00 © 2005 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on March 11, 2009 at 04:07 from IEEE Xplore. Restrictions apply.

Similarly, by considering the relation for the variance of random number of random i.i.d variables4, we have

terms of the probability of capture, i.e., it is desirable to select a power distribution that maximizes the probability of capture. In the following theorem, we address the effect of the power distribution on the throughput of the network. In particular, we demonstrate the impact of the mean and variance of the power distribution on the probability of capture.

M +1

Var[ P1 − γ

= Var[ P1 ] + γ 2 ( E[ M ]Var[ Pk ] + Var[ M ]( E[ Pk ])2 )

function f(P), respectively. Then, independent of f(p), the probability of capture is always bounded as follows:

2

+ [c 2 + c 3 ]µ − (2γN / G )c 3 µ + (γN / G) , c1 = 1 + (n − 1)qγ , c 3 = (1 − γ (n − 1)q) .

where

(4)

c 2 = (n − 1)q(1 − q)γ , 2

+ (n − 1) q (1 − q )µ 2 ] + {γN / G − [1 − γ (n − 1) q ]µ}2 ,

or equivalently, Pr{capture in a slot} ≤ nq[c1σ 2 + c 2 µ 2 ] / c1σ 2

+ [c 2 + c 3 2 ]µ 2 − (2γN / G )c 3 µ + (γN / G ) 2 , (14) In applying the Chebyshev’s inequality, we assume that the right-hand-side of equality (8) is non-negative. The latter assumption is true in almost all practical cases, as nq is usually equal to one (or quite close to one), and γ is significantly larger than one (typical values for γ ranges from 5 dB to 15 dB). ■

n

∑ Pr{node k capture s the channel }

(5 )

k =1

= n Pr{ node 1 captures the channel }

(6)

M +1

= nq Pr{GP1 /( N +

(13)

and

Proof. Since at most one user can capture the channel at each time slot, then we have Pr{capture in a slot } =

(12)

+ ( n − 1) q (1 − q)µ 2 ]}nq / σ 2 + γ 2 [(n − 1) qσ 2

2

2

(11)

2

Pr{capture in a slot } ≤ {σ 2 + γ 2 [( n − 1)qσ 2

2

2

2

By applying the one-tailed version of Chebyshev’s inequality5 to relation (8) and considering equalities (8) and (10), we conclude

σ 2 denote the mean and the variance of the power density

Pr{capture in a slot} ≤ nq[c1σ + c 2 µ ] / c1σ

2

= σ + γ [(n − 1)qσ + (n − 1)q (1 − q )µ ].

Theorem 1. A slotted Aloha medium access control with a randomized power control algorithm is considered. Let µ and

2

k

k =2

2

2

∑P ]

∑ GP ) ≥ γ}

(7 )

k

k =2

Based on equation (3), the following corollary directly follows from Theorem 1.

where equation (6) holds, since all the channel gains are assumed to be identical. Note that in relation (7), the number of interfering nodes (M) as well as the transmit power levels ( Pi ' s ) are random variables.

Corollary 1.1. Independent of the power density function f(P), the throughput of the network is always bounded as follows:

TH ≤ nq(1 − q)n −1 + nq[c1σ 2 + c2µ 2 ] / c1σ 2

It can be easily shown that equation (7) can be also written as follows.

+ [c2 + c32 ]µ 2 − (2γN / G)c3µ + (γN / G)2 , where c1, c2 , and c3 are defined similar to Theorem 1.

Pr{capture in a slot} =

(15)

M +1

nq Pr{P1 − γ

∑ P − [1 − γ(n − 1)q]µ ≥ γN / G − [1 − γ(n − 1)q]µ}.

(8)

k

k =2

Based on Wald’s lemma for the expected value of random number of random variables3, we have M +1

E[ P1 − γ

∑ P ] = E[P ] − γE[M ]E[P ]. k

1

k

(9 )

k =2

= [1 − γ (n − 1)q ]µ.

Theorem 1 provides a significant insight into the design of the power density function of the randomized power control algorithm. It can be easily shown that the derivative of the right-hand-side of relation (14) with respect to σ2 is always positive. The latter implies that for a fixed value of µ , (the upper bound for) the probability of capture increases as the 4

Let X1, X 2 ," be i.i.d random variables with E[ X 2 ] < ∞ . Let N be a

(10)

N

stopping variable,

and write

SN =

∑X

k .

If

E (N ) < ∞ ,

then

k =1

3

Wald’s Lemma: Let X1, X 2 ," be i.i.d random variables with E[ X ] < ∞ .

Var[ S N ] = E[ N ]Var[ X ] + Var[ N ]( E[ X ]) 2 [14].

N

Let N be a stopping variable, and write S N =

∑X

5

One-tailed version of Chebyshev’s inequality: For every random variable X

k . If E ( N ) < ∞ , then

2

and constant t ≥ 0 , we have Pr{ X − E[ X ] ≥ t} ≤ Var[ X ] /(Var[ X ] + t ) [14 ], which provides a tighter bound in comparison with the original Chebyshev’s inequality.

k =1

E[ S N ] = E[ X ]E[ N ] [14].

IEEE Communications Society / WCNC 2005

709

0-7803-8966-2/05/$20.00 © 2005 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on March 11, 2009 at 04:07 from IEEE Xplore. Restrictions apply.

variance of the power density function increases. This property is intuitively reasonable as the higher variance induces higher variation in the value of transmit power levels, which in turn increases the chance of capture at the receiver. Furthermore, Corollary 1.1 implies that one should select the power density function over the available power levels such that its mean and variance maximize the right-hand-side of relation (15) (see next section for illustrative examples). Now, let’s consider a more realistic model that the channel gains between different users and the central station may be different. This difference in channel gains may be due to different distances of the mobile users to the central station or due to dissimilar environmental circumstances. Under such a condition using identical power density function f(P) for all the users results in a higher probability of capture for the user nodes that have higher channel gains. To avoid such unfairness, we introduce the following shifting mechanism: A reference gain G (R) and a reference power density function f R ( P) (which can be designed based on Theorem 1 and Corollary 1.1) are given to all the user nodes in the network. Every node k (which is planning to transmit a packet in a slot) compares its associated channel gain Gk with the reference gain G ( R) . Node k shifts the reference power density function if f R ( P) to the right along the P-axis by G ( R) − Gk G ( R) − Gk ≥ 0 . Otherwise (i.e., if G ( R) − Gk < 0 ) node k shifts the reference power density function f R ( P) to the left along the P-axis by Gk − G ( R) . Since in relation (1) the gain Gk is a linear operator, the latter shifting mechanism causes the distribution of the received power of all the users at the central station to be identical. We note that there is a fundamental difference between the conventional power control operation in CDMA systems and our recommended approach for the slotted Aloha: in CDMA-based power control operation the received power of all the users (at the central station) are required to be equal. However, based on the shifting mechanism the distribution of the received power of all the users becomes identical. The Fair Randomized Power Control algorithm (FRPC) operates based on the following steps: Step 1. Based on the slotted Aloha mechanism, mobile node k plans to transmit a packet in the underlying time slot. Step 2. Node k compares its associated channel gain Gk (obtained by exchanging test packets) with the reference gain G ( R) . Step 3. If G ( R) − Gk ≥ 0 , node k shifts the reference power density function f R ( P) (designed based on Theorem 1 and Corollary 1.1) to the right along the P-axis by G ( R) − Gk . Otherwise, node k shifts f R ( P) to the left along the P-axis by Gk − G ( R ) .

IEEE Communications Society / WCNC 2005

710

Step 4. Node k randomly and independently selects a transmit power level based on the shifted power density function f k ( P) . Step 5. Node k transmits its packet under the selected power level in the underlying time slot. Node k returns to Step 1.

Under dynamic environments that the number of user nodes associated with the central station (n) may vary in time, the reference power density function f R ( P) should be modified adaptively (Corollary 1.1). This implies that the reference power density function can be defined as f R ( P, t ) , where t is a time parameter. f R ( P, t ) updates can be broadcasted from the central station in an event-based fashion. IV.

SIMULATION RESULTS

In our simulation environment 20 user nodes are located in a 50m x 50m square. The central station is positioned on the intersection of the diameters of the square. Noise power is set to be equal to -90 dBm and the path loss exponent ( α ) is assumed to be equal to 4. We assume that Gk = 1/ d k α , where d k is the distance of node k to the central station. The minimum required SINR is assumed to be 10dB. Every node is assumed to be immobile during the period of simulation and it transmits a packet in a slot with probability of q=0.05. In Fig. 1, we evaluate the throughput of the FRPC algorithm when only two power levels P1 = 1mW and P2 = 100mW are available. For this simulation, without loss of generality, we assume that all the nodes are in the same distance from the central station. Let q1 and q2 = 1 − q1 be the probability of transmission under power levels P1 and P2 , respectively. By substituting µ = P2 + ( P1 − P2 )q1 and σ 2 = ( P1 − P2 )2 q1 (1 − q1 ) in relation (15), and calculating the

derivative of the right-hand-side of (15) with respect to q1 , the optimal probability of transmission under power levels P1 and P2 can be calculated. For the underlying scenario the latter optimal values is equal to q1* = 0.28 and q2 * = 0.72 . We also evaluate the performance of the FRPC algorithm in an exhaustive manner as a function of various combinations of q1 and q2 . We observe a 27% throughput gain under the FRPC algorithm (under q1 * and q2 * ) with respect to classical slotted Aloha (with no power control). Note that under the slotted Aloha all transmissions are performed under 100 mW. We also note that the maximum achievable throughput under the given two power levels is about 0.5 (compared to 0.46 attained under the FRPC algorithm). Clearly, the latter minor difference is due to the fact that q1 * and q2 * maximize the upper bound of the throughput (and not necessarily the throughput itself).

0-7803-8966-2/05/$20.00 © 2005 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on March 11, 2009 at 04:07 from IEEE Xplore. Restrictions apply.

Fig. 2. Illustration of the throughput per node under slotted Aloha, RPC and FRPA.

Fig. 1. Illustration of throughput as a function of q1 under the FRPC algorithm and slotted Aloha.

In Fig. 2, we compare the fairness provided by the FRPC algorithm with those provided by the (conventional) Randomized Power Control Algorithm (RPC) and the classical slotted Aloha protocol (with no power control). For this simulation, we assume that the 20 nodes are randomly and uniformly distributed in the underlying square and a continuous spectrum of power between [1mW, 100mW] is available at each node. Similar to Fig. 1, under the slotted Aloha all transmissions are performed under 100 mW. As expected, slotted Aloha results in the highest variations in the throughput per node levels. Also, we observe that under the FRPC algorithm the values of throughput for different nodes have minor differences. In particular, we note that the minimum value of attained throughput among different nodes (i.e., MaxMin fairness) is highest under the FRPC algorithm. V.

[2] L. Roberts, “Aloha Packet System with and without Slots and Capture,” Comput. Commun. Rev., no. 5, 1975, pp. 28-42. [3] T. N. Saadawi and A. Ephremides, “Analysis, Stability, and Optimization of Slotted ALOHA with a Finite Number of Buffered Users,” IEEE Trans. Auto. Control, vol. AC-26, no. 3, June 1981, pp. 680-689. [4] A. B. MacKenzie and S. B. Wicker, “Stability of Multipacket Slotted ALOHA with Selfish Users and Perfect Information,” in Proc. IEEE INFOCOM, 2003. [5] A. Behzad and I. Rubin, “High Transmission Power Increases the Capacity of Ad Hoc Wireless Networks,“ to appear in IEEE Transactions on Wireless Communications. [6] S. A. Al-Semari and M. Guizani, “Channel Throughput of Slotted ALOHA in a Nakagimi Fading Environment,“ in Proc. IEEE ICC, 1997, pp. 605-609. [7] W. Luo and A. Ephremides, “Power Levels and Packet Lengths in Random Multiple Access,” IEEE Trans. Info. Theory, vol. 48, no. 1, January 2002, pp. 46-58.

CONCLUSIONS

In this paper, we develop and investigate a novel power control algorithm (the Fair Randomized Power Control algorithm; FRPC) to enhance the throughput level of the slotted Aloha medium access control mechanism. FRPC is developed based on the analysis of the capture effect for which the transmitter captures the channel only if the observed signal-to-interference and noise ratio (SINR) at receiver is above a certain threshold. Based on the one-tailed version of Chebyshev’s inequality, we derive a closed-form expression for an upper bound on the probability of capture (and an upper bound on the aggregate throughput) in the network. We then apply the latter result to the process of designing the FRPC algorithm for the slotted Aloha MAC mechanism.

[8] J. J. Metzner, “On Improving Utilization in ALOHA Networks,” IEEE Trans. Comm., vol. 24, no. 4, April 1976, pp. 447-448. [9] T. C. Hou and V. O. K. Li, “Transmission Range Control in Multihop Packet Radio Networks,” IEEE Trans. Commun., vol. 34, no. 1, January 1986, pp. 38-44. [10] R. O. LaMaiire, A. Krishna and M. Zorzi, ”On the Randomization of Transmitter Power Levels to Increase Throughput in Multiple Access Radio Systems,” Wireless Networks, vol. 4, 1998, pp. 263-277. [11] C. C. Lee, “Random Signal Levels for Channel Access in Packet Broadcast Networks,“ IEEE Jounal on Selected Areas in Communications, vol. 5, no. 6, July 1987, pp. 1026-1034. [12] H. Takanashi, H. Kayama, M. Iizuka and M. Morikura, “Enhanced Capture Effect for Slotted Aloha employing Transmission Power Control Corresponding to Offered Traffic,“ in Proc. IEEE ICC, 1998. [13] F. Berggren and J. Zander, “Throughput and Energy Consumption Tradeoffs in Pathgain-Based Constrained Power Control in Aloha Networks,” IEEE Communications Letters, vol. 4, no. 9, September 2000, pp. 283-285.

We are currently in the process of generalization of the derived results in this paper to mobile multihop wireless networks (with and without Rayleigh fading). Furthermore, the analysis of the energy efficiency of the Randomized Power Control algorithms is part of our ongoing research.

IEEE Communications Society / WCNC 2005

REFERENCES [1] N. Abramson, “The Aloha System – Another Alternative for Computer Communications,” in AFIPS Conference Proceedings, vol. 36, 1970, pp. 295-298.

[14] R. V. Hogg and A. T. Craig, Introduction to Mathematical Statistics, 1995, Prentice-Hall, Inc. [15] T. S. Rappaport. Wireless Communications: Principles and Practice. Second Edition, Prentice Hall, 2002.

711

0-7803-8966-2/05/$20.00 © 2005 IEEE

Authorized licensed use limited to: Univ of Calif Los Angeles. Downloaded on March 11, 2009 at 04:07 from IEEE Xplore. Restrictions apply.

On the performance of randomized power control ...

Mar 11, 2009 - advantage of the capture effect, under which the packet with the strongest ... In wireless communications, random access mechanisms are.

191KB Sizes 0 Downloads 198 Views

Recommend Documents

Impact of Power Control on the Performance of Ad Hoc ...
control (MAC) protocol such ,as time division multiple access. (TDMA), and a ..... receiver of more than one transmission at any time slot, a d ii) a node is not ...

Effects of Imperfect Power Control and Sectorization on ...
NF. F. FG. N. G. σµ γ. (21). By using (21), the user capacity x. N for singleclass CDMA in a reverse-link with imperfect power control and sectorization.

Research on Excitation Control of Flexible Power ... - IEEE Xplore
induction machine; direct-start; Back-to-back converters;. Speed control mode. I. INTRODUCTION. The power imbalance caused by power system fault.

Cluster-level Feedback Power Control for Performance ...
but consume more power in high-power states while having degraded functionality in ... control of power to adapt to a given power budget so that we reduce the power ..... where Tref is the time constant that specifies the speed of system response. ..

Inferring bounds on the performance of a control policy from a ... - ORBi
eralizations) of the following discrete-time optimal control problem arise quite frequently: a system, ... to high-enough cumulated rewards on the real system that is considered. In this paper, we thus focus on the evaluation of ... interactions with

Inferring bounds on the performance of a control policy ...
Dec 18, 2008 - and Reinforcement Learning (ADPRL 2009) conference proceedings published annually by IEEE ... How: many solutions. > Approximation ...

Inferring bounds on the performance of a control policy ...
Mar 16, 2009 - ))}l=1 n x t+1. =f (x t. ,u t. ) r t. =ρ(x t. ,u t. ) J. T h. (x. 0. )=∑ t=0. T1 ρ(x t. ,h(t,x ... Programming, Reinforcement Learning ... But. > Limited amount ...

Inferring bounds on the performance of a control policy from a ... - ORBi
The main philosophy behind the proof is the follow- ing. First, a sequence of .... Athena Scientific, Belmont, MA, 2nd edition, 2005. [2] D.P. Bertsekas and J.N. ...

Inferring bounds on the performance of a control policy ...
[3] R. Sutton and A. Barto, Reinforcement Learning, an. Introduction. MIT Press, 1998. [4] M. Lagoudakis and R. Parr, “Least-squares policy it- eration,” Jounal of Machine Learning Research, vol. 4, pp. 1107–1149, 2003. [5] D. Ernst, P. Geurts,

Efficient Power Control Based on Interference Range in ...
Energy management through power control in wireless ad hoc networks has a serious hidden ... Systems and Networks (EWSN) 2016. 15–17 February, Graz, ...

pdf-1874\high-performance-healthcare-using-the-power-of ...
... apps below to open or edit this item. pdf-1874\high-performance-healthcare-using-the-power-o ... ships-to-achieve-quality-efficiency-and-resilience.pdf.

A Review of Randomized Evaluation
Mar 2, 2007 - Miguel and Kremer [2004] investigated a biannual mass-treatment de-worming. [Research, 2000], a conditional cash transfer program in Nicaragua [Maluccio and Flores, 2005] and a conditional cash transfer program in Ecuador [Schady and Ar

Effects of sample size on the performance of ... -
area under the receiver operating characteristic curve (AUC). With decreasing ..... balances errors of commission (Anderson et al., 2002); (11) LIVES: based on ...

The Influence of Admixed Micelles on Corrosion Performance of ...
The Influence of Admixed Micelles on Corrosion Performance of reinforced mortar.pdf. The Influence of Admixed Micelles on Corrosion Performance of ...

The Effect of the Internet on Performance, Market ...
May 19, 2017 - are not the most popular ones, without affecting other movies. .... studies the impact of various policy, economic, and social changes, .... net users–where Internet users are people with access to the worldwide network. ..... on the

Evaluating the Impact of Reactivity on the Performance ...
interactive process to design systems more suited to user ... user clicks on a link or requests a Web page during its ses- sion. ...... Tpc-w e-commerce benchmark.

The effect of management structure on the performance ...
Mar 4, 2009 - procedure. In a multi-domain network a particular network management/controller may have complete information about its own domain but ...

On the Impact of Arousals on the Performance of Sleep and ... - Philips
Jul 7, 2013 - Electrical Engineering, Eindhoven University of Technology, Den Dolech. 2, 5612 AZ ... J. Foussier is with the Philips Chair for Medical Information ..... [6] J. Paquet, A. Kawinska, and J. Carrier, “Wake detection capacity of.

On the Impact of Arousals on the Performance of Sleep and ... - Philips
Jul 7, 2013 - techniques such as the analysis of the values in neighboring epochs [3] ..... Analysis Software (ADAS),” Physiology & Behavior, vol. 65, no. 4,.

On the performance of Outage Probability in Underlay ...
are exponential random variables. (RVs) with parameter ij. For example, the data link between. S and D denoted by ℎsd has the parameter sd. To facilitate.

The impacts of performance pay on teacher ...
Relative to these papers, our paper aims to contribute by providing a direct ..... does not rely on within-teacher variation in treatment and makes no attempt to ..... significantly less likely to assign “Worksheets” and are more likely to assign

On the Complexity and Performance of Parsing with ... - GitHub
seconds to parse only 31 lines of Python. ... Once these are fixed, PWD's performance improves to match that of other ...... usr/ftp/scan/CMU-CS-68-earley.pdf.