Optimal Base Station Density for Power Efficiency in Cellular Networks


 Randell Gilmore
 1 years ago
 Views:
Transcription
1 Optimal Base Station Density for Power Efficiency in Cellular Networks Sanglap Sarkar, Radha Krishna Ganti Dept. of Electrical Engineering Indian Institute of Technology Chennai, 636, India {ee11s53, Martin Haenggi Dept. of Electrical Engineering University of Notre Dame Notre Dame, IN 46556, USA Astract In cellular networks, cell size reduction is an important technique for improving the spectral reuse and achieving higher data rates. In addition, it results in power savings as it leads to a decrease in transmit power. However, it is not clear if the transmit power can e indefinitely decreased with the cell sizes. In this paper, we analyze the impact of transmit power reduction (cell size reduction) on the performance of the network. More precisely, we otain a lower ound on the transmit power such that a minimum coverage and a minimum data rate can e guaranteed. We then analyze the area power consumption metric, which denotes the total power consumed per unit area. Under the constraints of target coverage and target data rate, the area power consumption is minimized and the optimal ase station density is otained. For a path loss exponent α > 4, we oserve the existence of a minimum cell size elow which shrinking the cell would result in an overall increase of power. However, for α 4, there exists no such optimal cellsize, as the area power consumption increases with ase station density. Index Terms Small cells, green communication, cell size, quality of service, power consumption, power efficiency, optimal ase station density. I. INTRODUCTION Cell size reduction provides increased spectral reuse and increased data rates to moile users. As the cell size decreases, the numer of users per ase station (BS) decreases leading to a greater andwidth (or time share) per user. Cell size reduction can e achieved y increasing the density of the ase stations, either y increasing the numer of macro ase stations or adding tiers of low powered ase stations. There are two advantages of cell size reduction. Firstly, increased andwidth per user. Secondly, lower transmit power since the moile user is much closer to a ase station. In this paper, we investigate if the downlink transmit power can e decreased aritrarily y increasing the density of ase stations for a given target rate and coverage. It turns out that after a certain power threshold, noise plays a significant role on oth coverage and rate. For α > 4, we otain an expression for the optimum ase station density which minimizes area power consumption and maximizes power efficiency 1 under target rate and coverage constraints. If the cell density exceeds an optimal threshold 1 Power efficiency is defined as inverse of the area power consumption. We call the network to e power efficient if the area power consumption decreases with increase of ase station density. the total power consumption increases. The optimum density gives us a limit up to which cell size can e shrunk and power savings can e achieved. It also indicates when operator should stop further deployment of small cells in this regime. For α 4, we oserve that increasing the ase station density leads to an overall increase of power. II. SYSTEM MODEL AND PROBLEM FORMULATION A. System Model In this paper, we assume a single tier of BSs and focus on the downlink. The locations of the ase stations are modeled y a spatial Poisson point process (PPP) [1] Φ of density. We assume that users forms a stationary point process with density λ u that is independent of the ase station process. Since the users are uniformly distriuted, the average numer of users per cell is λ u /. We assume a path loss function l(x) = K x α, α > 2, where K is a constant chosen 2 as Without loss of generality, we focus on a moile user at the origin for computing coverage and rate. We assume that each moile user is served y its closest BS. We also assume that all BSs transmit with the same power P. The andwidth (BW) per user is a random variale that depends on the cell size and the numer of users in the cell. However, for ease of analysis we set it to e equal to the total availale andwidth (B) divided y the average numer of users, i.e., B u ( ) = B λ u. Hence the average rate per user is R = B u ( )E[ln(1 + SINR)]. Since the user BW depends on the densities, the noise power is F kt B u ( ), where F is the receiver noise figure, k is the Boltzmann constant, and T is the amient temperature. The small scale fading is denoted y h x which is assumed to follow an exponential distriution (square of Rayleigh fading) with unit mean. The signaltointerferenceplusnoise ratio for 2 A typical femtocell covers a range of 4 m with a transmit power of.2 W [2]. The received power at the cell oundary should e at least equal to the receiver sensitivity, which is 8 dbm. So r = 4, α = 3, transmit power P t =.2, received power P r = = K P t r α. Hence, K =
2 the moile user at the origin is h x x α SINR = σ 2 P + y Φ \{x} h y y, α where σ 2 = F kt B/(λ u K), and x Φ is the BS closest to the origin. The coverage proaility is defined as (P ) = P(SINR > θ), and the spectral efficiency is defined as τ(p ) = E[ln(1 + SINR)]. B. Prolem Formulation With BSs transmitting at power P, we define the area transmit power consumption as P. In addition to the transmission power, each BS requires a fixed amount of operational power which we denote y P. The power P includes the fixed operational power of the hardware and also the transmit power of the pilot tones. Hence the area power consumption is (P + P ). When the transmit power is infinity, it is shown in [3] that the SINR distriution and hence the coverage proaility and spectral efficiency does not depend on the BS density. The rate demanded y each user is denoted y ; it is independent of the ase station density. The interferencelimited spectral efficiency, corresponding to P =, is τ( ). It is independent of the ase station density and depends only on path loss exponent α. So, irrespective of the transmit power, the maximum achievale rate for a particular density is given y τ( )B u ( ). The user demand rate can e satisfied only if it is less than the maximum achievale rate. But if the user demand exceeds the maximum achievale rate, then the est the operator can do is to support a data rate of τ( )B u ( ). So, the operator must satisfy the target rate or maximum achievale rate, whichever is minimum, with a maximum rate outage δ. The limiting coverage proaility that can e achieved for a given SINR threshold θ and path loss exponent α is given y ( ). So, the operator must provide ( ) coverage with a maximum outage of ɛ. Our goal is to otain the optimal parameters λ and transmit power P so that the area power consumption (P + P ) is minimal while maintaining the QoS constraints of rate and coverage. Formally, λ = arg min { (P + P )} s.t. (P ) (1 ɛ) ( ), (1) R(P ) (1 δ) min{, τ( )B u ( )}. (2) We consider oth the rate and coverage constraints for the following two reasons: Even though B u ( ) ln(1 + SINR) might e high, SINR might e low. This might e ecause of a large BW B. Because of the inclusion of, the lower ound on transmit power otained from the two constraints, may scale differently with the BS density as can e seen in Lemma 1 and Lemma 2 and Fig. 3 and Fig. 6. C. Prior Work Energy efficiency in HetNets has received increasing attention recently. Mostly only interferencelimited networks have een considered [4], [5]. Different energy saving techniques in communication networks, have een presented in [6]. Energy efficiency in cellular networks has also een analyzed in [7],[8]. However, the authors have focused on a hexagonal grid model with fixed downlink transmit power. In [4], an optimal asestation density for oth homogeneous and heterogeneous cellular networks has een otained under the constraint of target rate. However, noise has een ignored. Coverage constraints have not een considered, either.energy efficiency in HetNets has een studied in [5], which provides an optimal macropico density ratio that maximizes the overall energy efficiency under fixed noise assumption (i.e., noise power does not vary with the allocated andwidth). In this paper, we consider noise and oth target coverage constraints and target rate constraints. We also show how the transmit power has to e scaled down with increase of ase station density. III. OPTIMAL POWER FOARGET COVERAGE AND RATE A. Minimum transmit power for coverage As the BS density increases, the transmit power of the ase stations may e decreased ecause of the decreasing cell size. However, reducing the transmit power, decreases the coverage proaility ecause of the noise. See Fig. 1. In the next lemma we evaluate the transmit power required to achieve (1 ɛ) ( ) coverage. Lemma 1. The minimum downlink transmit power for which a minimum SINR of θ can e guaranteed for a fraction 1 ɛ of the users is given y where k 1 = P c k 1 θσ2 Γ(α/2+1) ɛπ α/2 [1+ρ(θ,α)] α/2. λ α/2 1, (3) Proof: The coverage proaility without noise (or infinite power) is [3] ( ) = (1 + ρ(θ, α)) 1, (4) where ρ(θ, α) = θ 2/α 1 dx. The coverage proaility 1+x α/2 with noise is [3] (P ) = π e π(1+ρ(θ,α))v θσ 2 v α/2 P 1 dv. (5) Now, using the sustitution θσ 2 P 1 s in (5) and the approximation e svα/2 (1 sv α/2 ) for small s (or, equivalently, small σ 2 ), we get ( θσ 2 ) Γ(α/2 + 1) (P ) ( ) 1, (6) P λ α/2 1 [π(ρ(θ, α) + 1)] α/2 where Γ(x) = t x 1 e t dt is the standard gamma function. This approximation (linearization of the exponential term) is
3 1 Coverage proaility vs Transmit power Min transmit power for target coverage vs ase station density 7 Coverage proaility: (P) =1/1 2 m 2, θ= 15dB =1/5 2 m 2, θ= 15dB =1/4 2 m 2, θ= 15dB ( )(θ= 15dB, α=4) =1/1 2 m 2, θ= db =1/5 2 m 2, θ=db =1/4 2 m 2, θ=db ( )(θ=db, α=4) Transmit power: P (dbm) Fig. 1: Effect of transmit power on coverage proaility: User density λ u =.1 m 2, α = 4 and σ 2 = W. min power for target coverage: P c * (dbm) θ=5db, α=3 θ=2db, α=3 θ=5db, α=4 θ=2db, α= Base station density: λ (in m 2 ) 1 2 Fig. 3: Minimum transmit power for target coverage: User density λ u =.1 m 2, Outage proaility ɛ =.25 and σ 2 = W. Tightness of approximation used to evaluate coverage proaility Rate vs transmit power Coverage Proaility: (P) With Approximation for =1 6 m 2 Without Approximation for =1 6 m 2 With Approximation for =1 2 m 2 Without Approximation for =1 2 m Transmit Power P (dbm) Fig. 2: Tightness of numerical approximation e (θσ2 P 1 )v α/2 (1 θσ 2 P 1 v α/2 ) used to evaluate coverage proaility: SINR threshold θ = 3dB, User density λ u =.1 m 2, α = 3 and σ 2 = W. Rate: R (in ps) = 1/1 2 m 2 = 1/5 2 m 2 = 1/2 2 m 2 = 1/7 2 m 2 R 1 T = 4Kps Transmit power: P (dbm) Fig. 4: Effect of noise on rate: User density λ u =.1 m 2, α = 4.5 and σ 2 = W. very tight, as can e seen from Fig. 2. The lower ound Pc on transmit power P can e otained y comining (4) and (6) with the coverage constraint (1). With increasing ase station density, the operator can scale down the transmit power at a rate proportional to λ 1 α/2 (as shown in Fig. 3), while guaranteeing a certain minimum coverage. B. Minimum transmit power for target data rate The effect of noise on rate is shown in Fig. 4. The transmit power can decrease with increasing BS density. However, a minimum transmit power is required to comat the noise and provide average rate to satisfy the user demand or the maximum achievale rate, whichever is minimum. This constraint imposes a limit on the downlink transmit power, which is provided in the next lemma. Lemma 2. The minimum downlink transmit power that achieves a fraction 1 δ of the minimum of the target rate and maximum supported rate τ( )B u ( ) is given y Pr min{h(), k 2 }, (7) λ α/2 1
4 Spectral efficiency: τ (in ps/hz) Tightness of approximation used to evaluate spectral efficiency without approx, =1 3 m 2, α=4 with approx, =1 3 m 2, α=4 without approx, =1 4 m 2, α=4 with approx, =1 4 m 2, α=4 without approx, =1 4 m 2, α=3 with approx, =1 4 m 2, α= Transmit power: P (in dbm) Fig. 5: Tightness of numerical approximation e σ2 P rα (e t 1) [1 σ2 P rα (e t 1)] used to evaluate spectral efficiency: User density λ u =.1m 2, noise power σ 2 = W. Min power for target rate: P r * ( in dbm ) Min transmit power target rate vs ase station density =1Kps, α=3 =1Mps, α=3 =1Gps, α=3 =1Kps, α=4 =1Mps, α=4 =1Gps, α= Base station density λ (m 2 ) Fig. 6: Minimum transmit power for target rate: User density λ u =.1 m 2, total andwidth B = 2 MHz, Outage proaility δ =.25 and noise power σ 2 = W. where k 2 = σ2 g(α) δτ( ), h() = g(α) = π α/2 t> z> e z z α/2 σ 2 g(α) [τ( ) (1 δ){ λu B }], and (e t 1)e z{(et 1) 2/α (e t 1) 2/α 1 1+x α/2 dx} dtdz. Proof: The spectral efficiency τ(p ) has een derived in [3] as τ(p ) E[ln(1 + SINR)] = e πr 2 P r> t> e σ 2 λ rα (e t 1) L Ir (r α (e t 1))dt2π rdr, where L Ir (r α (e t 1)) is the Laplace transform of the interference I r evaluated at r α (e t 1), which is given y ( ) exp π r 2 (e t 1) 2/α 1 dx. (e t 1) 1 + x 2/α α/2 Since σ 2 is very small, we use the approximation e σ2 P rα (e t 1) (1 σ2 P rα (e t 1)) in (8). This approximation is tight as shown in Fig. 5. Sustituting π r 2 z in (8), we otain τ(p ) = τ( ) σ 2 g(α)p 1 λ α/2+1 + o(σ 2 ), σ 2, (9) where τ( ) is the spectral efficiency at zero noise or high power, which is independent of the BS density [3]. A fraction 1 δ of the users must achieve a data rate at least equal to the target rate or maximum achievale rate, whichever is (8) Rate ( in Mps ) Target rate and maximum acheivale rate Vs ase station density Maximum achievale rate Target rate or user demand Base station density: ( in m 2 ) Fig. 7: Maximum achievale rate and target rate (user demand): User density λ u =.1 m 2, α = 4, total andwidth B = 2 MHz and target rate (user demand) = 5 kps. minimum. This, in turn, imposes a constraint on the spectral efficiency τ(p ): { } RT τ(p ) (1 δ) min B u ( ), τ( ). (1) Comining (1) and (9) we get a lower ound P r on the transmit power P. Depending on the target data rate, the transmit power of the ase stations can e scaled down (under target data rate constraint) with the increase of small cell density, at a rate proportional to λ 1 α 2 or faster than that as shown in Fig. 6. IV. OPTIMUM BASE STATION DENSITY To ensure QoS (oth coverage and user demand rate), the minimum downlink transmit power should e P m=max(p c, P r ). So, the total power consumption y a ase station is P m + P. The area power consumption follows as Λ 1 Λ 2
5 P A ( ) = (P m + P ) = (max(p c, P r ) + P ). Using (3) and (7), we get P A ( ) = max{k 1, min{h( ), k 2 }} λ α/2 2 + P. (11) The optimal λ is then otained y setting dp A d =. Theorem 1. When α > 4, the optimum ase station density λ that minimizes the area power consumption under coverage and rate constraints is given y where and [ (α 4) max{k1, k 2 } ] 2 λ = 2P [ (α 4)k1 ] 2 2P R 2, R 1, [ (α 4)(Bτ( )) α/2 1 ] 2 k 1 R 1 =, 2P λ α/2 1 u [ (α 4)(Bτ( )) α/2 1 ] 2 max{k 1, k 2 } R 2 =. 2P λ α/2 1 u Proof: We egin the proof y considering 3 different ranges of. We define Λ 1 = λ u Bτ( ) and Λ k 2 = 1c 2 k 1τ( ) c 1. λ We also rewrite h( ) as h( ) = c 1 τ( ) c 2. The demand rate is at least equal to the maximum achievale rate B u ( )τ( ) only when (, Λ 1 ], as can e seen in Fig. 7. Rearranging the expressions of h( ), Λ 1, Λ 2 and k 2 we get k 2 (, Λ 1 ], min{h( ), k 2 } = h( ) (Λ 1, Λ 2 ), h( ) [Λ 2, ). Rearranging the expressions of Λ 2, k 1 and h( ), it can e shown that, k 1 h( ) only for [Λ 2, ). Therefore, the area power consumption given in (11) can e rewritten as max{k 1, k 2 } + P λ α/2 2 (, Λ 1 ], h( ) P A = + P (Λ 1, Λ 2 ), (12) λ α/2 2 k 1 + P λ α/2 2 [Λ 2, ). Area power consumption: P A (in W/m 2 ) Area power consumption: analytical and simulation results α=6 α=5 α=4 α= Base station density: λ (in m 2 ) 1 Fig. 8: Area power consumption: Target rate = 5Mps, SINR threshold θ = 12 db, User density λ u =.1 m 2, Outage proaility ɛ =.25, Rate outage δ =.25, P = 1.5 W and noise power σ 2 = W. The dashed lines correspond to our analysis while the old lines are otained y Monte Carlo simulation. Rearranging the expressions of R 1, R 2, Λ 1, Λ 2 and using the When R 1 < < R 2, the optimal ase station density λ is derivative of (12), for α > 4, we otain the solution to the following equation dp A (Λ 1 ) λ α/2 2 (2P c 2 2) λ α/2 1 (4P τ( )c 2 ) + 2λ α/2 P (τ( )) 2 R 2, d dp A (Λ 1 ) λ <, dp A(Λ 2 ) > R ((α 4)c 1 τ( )) + ((α 6)c 1 c 2 ) =, 1 < < R 2, d d dp where c 1 = σ 2 g(α), c 2 = (1 δ) λ u A (Λ 2 ) R B. There exists no T R 1. d optimum density for α 4. Therefore for α > 4, (, Λ 1 ] R 2, λ (Λ 1, Λ 2 ) R 1 < < R 2, (13) [Λ 2, ) R 1. Now, for α > 4, optimal ase station density λ can e otained y setting the derivative of (12) to zero. When α 4, the total area transmit power consumption Pm as well as the total area fixed power consumption P increase with ase station density, for (, Λ 1 ] and [Λ 2, ), see (12). Hence, in this case, there exists no optimal density. The area power consumption always increases with ase station density for α = 3 and α = 4 as can e seen in Fig. 8. The simulation curves are otained using the Monte Carlo method. This result indicates that cell size shrinking or dense small cell deployment is not powerefficient for these values of the path loss exponent. Fig. 8 also shows the area power consumption and the optimal ase station densities for α = 5 and α = 6. This result indicates that small cell deployment can e power efficient in uran areas (where it is possile that α > 4).
6 V. CONCLUSION In this paper, we have derived a lower ound on downlink transmit power in a homogeneous Poisson networks under target coverage and target data rate constraints. For a path loss exponent α > 4, we have proven the existence of an optimal deployment density after which further reduction of the cell size is not efficient from an energy viewpoint. Our results indicate that, for α 4, counterintuitively, smaller BS densities lead to improved area power consumption. In other words, increasing the BS density is not green for α 4. REFERENCES [1] M. Haenggi, Stochastic geometry for wireless networks. Camridge University Press, 212. [2] V. Chandrasekhar, J. Andrews, and A. Gatherer, Femtocell networks: a survey, Communications Magazine, IEEE, vol. 46, no. 9, pp , 28. [3] J. G. Andrews, F. Baccelli, and R. K. Ganti, A tractale approach to coverage and rate in cellular networks, IEEE Trans. on Communications, vol. 59, no. 11, pp , Nov [4] D. Cao, S. Zhou, and Z. Niu, Optimal ase station density for energyefficient heterogeneous cellular networks, International Conference on Communications (ICC), IEEE, pp , 212. [5] T. Q. Quek, W. C. Cheung, and M. Kountouris, Energy efficiency analysis of twotier heterogeneous networks, in 11th European Wireless Conference 211Sustainale Wireless Technologies (European Wireless). VDE, 211, pp [6] X. Wang, A. V. Vasilakos, M. Chen, Y. Liu, and T. T. Kwon, A survey of green moile networks: Opportunities and challenges, Moile Networks and Applications, vol. 17, no. 1, pp. 4 2, 212. [7] F. Richter and G. Fettweis, Cellular moile network densification utilizing micro ase stations, in International Conference on Communications (ICC). IEEE, 21, pp [8] F. Richter, A. J. Fehske, P. Marsch, and G. P. Fettweis, Traffic demand and energy efficiency in heterogeneous cellular moile radio networks, in 71st Vehicular Technology Conference (VTC). IEEE, 21Spring, pp. 1 6.
HighRate Codes That Are Linear in Space and Time
1804 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL 48, NO 7, JULY 2002 HighRate Codes That Are Linear in Space and Time Babak Hassibi and Bertrand M Hochwald Abstract Multipleantenna systems that operate
More informationSubspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity
Subspace Pursuit for Compressive Sensing: Closing the Gap Between Performance and Complexity Wei Dai and Olgica Milenkovic Department of Electrical and Computer Engineering University of Illinois at UrbanaChampaign
More informationApproximately Detecting Duplicates for Streaming Data using Stable Bloom Filters
Approximately Detecting Duplicates for Streaming Data using Stable Bloom Filters Fan Deng University of Alberta fandeng@cs.ualberta.ca Davood Rafiei University of Alberta drafiei@cs.ualberta.ca ABSTRACT
More informationHIGH DIMENSIONAL GRAPHS AND VARIABLE SELECTION WITH THE LASSO. By Nicolai Meinshausen and Peter Bühlmann ETH Zürich
The Annals of Statistics HIGH DIMENSIONAL GRAPHS AND VARIABLE SELECTION WITH THE LASSO By Nicolai Meinshausen and Peter Bühlmann ETH Zürich The pattern of zero entries in the inverse covariance matrix
More informationTHE PROBLEM OF finding localized energy solutions
600 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 45, NO. 3, MARCH 1997 Sparse Signal Reconstruction from Limited Data Using FOCUSS: A Reweighted Minimum Norm Algorithm Irina F. Gorodnitsky, Member, IEEE,
More informationTurning 802.11 InsideOut
Turning 802.11 InsideOut Pravin Bhagwat Bhaskaran Raman Dheeraj Sanghi + Abstract The past decade has seen communication revolution in the form of cellular telephony as well as the Internet, but much
More informationEnergy Efficient Schemes for Wireless Sensor Networks with Multiple Mobile Base Stations
Energy Efficient Schemes for Wireless Sensor Networks with Multiple Mobile s Shashidhar Rao Gandham, Milind Dawande, Ravi Prakash and S. Venkatesan Department of Computer Science School of Management University
More informationWhich Free Lunch Would You Like Today, Sir?: Delta Hedging, Volatility Arbitrage and Optimal Portfolios
Which Free Lunch Would You Like Today, Sir?: Delta Hedging, Volatility Arbitrage and Optimal Portfolios Riaz Ahmad Course Director for CQF, 7city, London Paul Wilmott Wilmott Associates, London Abstract:
More informationEstimators Also Need Shared Values to Grow Together
TECHNICAL REPORT TR1104, COMNET, TECHNION, ISRAEL 1 Estimators Also Need Shared Values to Grow Together Erez Tsidon, Iddo Hanniel and Isaac Keslassy Technion Qualcomm {erezts@tx., ihanniel@, isaac@ee.}technion.ac.il
More informationHow to Use Expert Advice
NICOLÒ CESABIANCHI Università di Milano, Milan, Italy YOAV FREUND AT&T Labs, Florham Park, New Jersey DAVID HAUSSLER AND DAVID P. HELMBOLD University of California, Santa Cruz, Santa Cruz, California
More informationIN just the past year, preliminary interest and discussions
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 32, NO. 6, JUNE 2014 1065 What Will 5G Be? Jeffrey G. Andrews, Fellow, IEEE, Stefano Buzzi, Senior Member, IEEE, Wan Choi, Senior Member, IEEE, Stephen
More informationFoundations of Data Science 1
Foundations of Data Science John Hopcroft Ravindran Kannan Version /4/204 These notes are a first draft of a book being written by Hopcroft and Kannan and in many places are incomplete. However, the notes
More informationNoncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas
3590 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO., NOVEMBER 200 Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas Thomas L. Marzetta Abstract A cellular base
More informationRECENTLY, there has been a great deal of interest in
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 47, NO. 1, JANUARY 1999 187 An Affine Scaling Methodology for Best Basis Selection Bhaskar D. Rao, Senior Member, IEEE, Kenneth KreutzDelgado, Senior Member,
More informationHow Bad is Forming Your Own Opinion?
How Bad is Forming Your Own Opinion? David Bindel Jon Kleinberg Sigal Oren August, 0 Abstract A longstanding line of work in economic theory has studied models by which a group of people in a social network,
More informationSteering User Behavior with Badges
Steering User Behavior with Badges Ashton Anderson Daniel Huttenlocher Jon Kleinberg Jure Leskovec Stanford University Cornell University Cornell University Stanford University ashton@cs.stanford.edu {dph,
More informationSelling assets: When is the whole worth more than the sum of its parts?
Selling assets: When is the whole worth more than the sum of its parts? Robert Marquez University of California, Davis Rajdeep Singh University of Minnesota October, 214 Abstract When is it better to sell
More informationThe SmallWorld Phenomenon: An Algorithmic Perspective
The SmallWorld Phenomenon: An Algorithmic Perspective Jon Kleinberg Abstract Long a matter of folklore, the smallworld phenomenon the principle that we are all linked by short chains of acquaintances
More informationRealTime Dynamic Voltage Scaling for LowPower Embedded Operating Systems
RealTime Dynamic Voltage Scaling for LowPower Embedded Operating Syste Padmanabhan Pillai and Kang G. Shin RealTime Computing Laboratory Department of Electrical Engineering and Computer Science The
More informationObject Detection with Discriminatively Trained Part Based Models
1 Object Detection with Discriminatively Trained Part Based Models Pedro F. Felzenszwalb, Ross B. Girshick, David McAllester and Deva Ramanan Abstract We describe an object detection system based on mixtures
More informationNearOptimal Sensor Placements in Gaussian Processes
Carlos Guestrin Andreas Krause Ajit Paul Singh School of Computer Science, Carnegie Mellon University GUESTRIN@CS.CMU.EDU KRAUSEA@CS.CMU.EDU AJIT@CS.CMU.EDU Abstract When monitoring spatial phenomena,
More informationRevisiting the Edge of Chaos: Evolving Cellular Automata to Perform Computations
Revisiting the Edge of Chaos: Evolving Cellular Automata to Perform Computations Melanie Mitchell 1, Peter T. Hraber 1, and James P. Crutchfield 2 In Complex Systems, 7:8913, 1993 Abstract We present
More informationThe Distributional Impact of Public Services When Needs Differ
DISCUSSION PAPER SERIES IZA DP No. 4826 The Distributional Impact of Public Services When Needs Differ Rolf Aaberge Manudeep Bhuller Audun Langørgen Magne Mogstad March 2010 Forschungsinstitut zur Zukunft
More informationFast Solution of l 1 norm Minimization Problems When the Solution May be Sparse
Fast Solution of l 1 norm Minimization Problems When the Solution May be Sparse David L. Donoho and Yaakov Tsaig October 6 Abstract The minimum l 1 norm solution to an underdetermined system of linear
More informationTruthful Mechanisms for OneParameter Agents
Truthful Mechanisms for OneParameter Agents Aaron Archer Éva Tardos y Abstract In this paper, we show how to design truthful (dominant strategy) mechanisms for several combinatorial problems where each
More informationMonte Carlo methods in PageRank computation: When one iteration is sufficient
Monte Carlo methods in PageRank computation: When one iteration is sufficient K.Avrachenkov, N. Litvak, D. Nemirovsky, N. Osipova Abstract PageRank is one of the principle criteria according to which Google
More informationCompressed Sensing with Coherent and Redundant Dictionaries
Compressed Sensing with Coherent and Redundant Dictionaries Emmanuel J. Candès 1, Yonina C. Eldar 2, Deanna Needell 1, and Paige Randall 3 1 Departments of Mathematics and Statistics, Stanford University,
More informationAgilent Time Domain Analysis Using a Network Analyzer
Agilent Time Domain Analysis Using a Network Analyzer Application Note 128712 0.0 0.045 0.6 0.035 Cable S(1,1) 0.4 0.2 Cable S(1,1) 0.025 0.015 0.005 0.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Frequency (GHz) 0.005
More informationA PROCEDURE TO DEVELOP METRICS FOR CURRENCY AND ITS APPLICATION IN CRM
A PROCEDURE TO DEVELOP METRICS FOR CURRENCY AND ITS APPLICATION IN CRM 1 B. HEINRICH Department of Information Systems, University of Innsbruck, Austria, M. KAISER Department of Information Systems Engineering
More informationThe Capital Asset Pricing Model: Some Empirical Tests
The Capital Asset Pricing Model: Some Empirical Tests Fischer Black* Deceased Michael C. Jensen Harvard Business School MJensen@hbs.edu and Myron Scholes Stanford University  Graduate School of Business
More information