Radio planning problems and methodologies in wireless networks. Multiple access schemes

Transcription

1 Politecnico di Milano Advanced Network Technologies Laboratory Radio planning problems and methodologies in wireless networks. Multiple access schemes Prof. Antonio Capone Master program Communication Technologies, Systems and Networks Universidad Politécnica de Valencia May

2 Summary Introduction to wireless networks Main differences with wired networks Radio channel Mobility management Multiple access Physical channels (FDMA, TDMA, CDMA) Packet access (scheduled access, random access) Radio resource planning Coverage planning Frequency assignment Joint coverage and capacity planning Antonio Capone: Wireless Networks 2

3 Wireless Networks Wireless or or wired, what is is better? Well, it it depends on on the situation! wireless wired Is the transmission medium the only difference? The peculiar medium characteristics have great impact on system characteristics Wireless networks allow users to move and naturally manage mobility Antonio Capone: Wireless Networks 3

4 Network architecture Backbone network Access network Antonio Capone: Wireless Networks 4

5 Wireless access networks Wireless networks are mainly access networks Backbone networks composed of radio point-to-point links are usually not considered wireless networks Wireless access networks are more challenging and have many fundamental differences with respect to wired access networks The first main difference is that the transmission medium is broadcast Antonio Capone: Wireless Networks 5

6 Broadcast channel Centralized broadcast channel Distributed broadcast channel Antonio Capone: Wireless Networks 6

7 Centralized broadcast channel Fixed access point (cellular systems, WLAN, WMAN) Wired network Mobile-access point connection Antonio Capone: Wireless Networks 7

8 Centralized broadcast channel Cellular coverage: The territory coverage is obtained by Base Stations BS (or Access Points) that provide radio access to Mobile Stations MS within a service area called CELL Base Station Mobile Station Cell Antonio Capone: Wireless Networks 8

9 Distributed broadcast channel Ad-hoc wireless networks (mesh networks, sensor networks) mobile- mobile connections Antonio Capone: Wireless Networks 9

10 Distributed broadcast channel In multi-hop operation mobile stations can forward information source relay relay destination Antonio Capone: Wireless Networks

11 Wired-Wireless networks: Main differences Shared transmission medium Multiple access mechanisms Radio resource reuse Central swith cable Radio channel Antonio Capone: Wireless Networks

12 Wired-Wireless networks: Main differences Radio channel Variable channel characteristics Advanced modulation and coding schemes Antonio Capone: Wireless Networks 2

13 Wired-Wireless networks: Main differences User mobility Stand-by mobility Active session (conversation) mobility Antonio Capone: Wireless Networks 3

14 Wired-Wireless networks: Main differences In this course we ll focus on the issues related to the shared wireless medium: Multiple access (Part A) Resource planning and reuse (Part B) But first a few comments on: Wireless channel Mobility management Antonio Capone: Wireless Networks 4

15 Wireless channel Very bad channel compared to other wired mediums Signals propagation is subject to : High attenuation due to distance Supplementary attenuation due to obstacles Multipath propagation Antonio Capone: Wireless Networks 5

16 Wireless channel: radio spectrum Radio waves Wave length Light speed Frequency f λ = c f c = 3 8 m/s s( t) = cos(2π ft + ϕ) Antonio Capone: Wireless Networks 6

17 Wireless channel: radio spectrum Antonio Capone: Wireless Networks 7

18 Wireless channel: radio spectrum Higher frequencies: more bandwidth less crowded spectrum but greater attenuation through walls Lower frequencies bandwidth limited longer antennas required greater antenna separation required several sources of man-made noise Antonio Capone: Wireless Networks 8

19 Wireless channel: antennas Transmission and reception are achieved by means of an antenna An antenna is an electrical conductor or system of conductors Transmission - radiates electromagnetic energy into space Reception - collects electromagnetic energy from space In two-way communication, the same antenna can be used for transmission and reception Isotropic antenna (idealized) Radiates power equally in all directions (3D) Real antennas always have directive effects (vertically and/or horizontally) Antonio Capone: Wireless Networks 9

20 Wireless channel: attenuation Isotropic radiator transmitting power P T uniformly in all directions distance d source area Power density F at distance d is: PT 2 F = [W/m ] 2 4πd Antonio Capone: Wireless Networks 2

21 Wireless channel: attenuation Directive characteristics of real antennas concentrate power in some directions This effect can be modeled using the gain g(θ) in the direction θ g( θ ) = power at distance d in direction θ 2 P / 4πd The maximum gain g T is convetionally in the direction θ =. T Antonio Capone: Wireless Networks 2

22 Wireless channel: attenuation The power density in the maximum gain direction is given by: F( d) = PT g 4πd T 2 [W/m The product P T g T is called EIRP (Effective Isotropically Radiated Power) and it is the power required to reach the same power density with an isotropic radiator 2 ] Antonio Capone: Wireless Networks 22

23 Wireless channel: attenuation The received power depends on the power density at the receiver antenna and its equivalent area: P = F( d) R A e For an isotropic antenna we have: While for a directive antenna we can concentrate energy: P = F( d) g R R A e A e = 2 λ 4π Where g R is the receiver antenna gain 2 Therefore: P R = P λ 4πd Antonio Capone: Wireless Networks 23 T g T g R

24 Antonio Capone: Wireless Networks 24 Wireless channel: free space model (Friis) The received power is d -2 This is known as the free space propagation model It can be used for example with point-topoint radio links = = fd c g g P d g g P P R T T R T T R π π λ

25 Wireless channel: propagation impairments Unfortunately, in real environments the propagation of electromagnetic waves is more complex that in free space: Reflection Shadowing Antonio Capone: Wireless Networks 25

26 Wireless channel: propagation impairments Diffraction When the surface encountered has sharp edges. Bending the wave Scattering When the wave encounters objects smaller than the wavelength (vegetation, clouds, street signs) Antonio Capone: Wireless Networks 26

27 Wireless channel: two ray model Just in the case of reflection with only two rays (a direct ray and a reflected one) received power is quite different wrt free space 2 2 h h It can shown that: T R d P R ( d) P T g T g R d 4 h T h R Antonio Capone: Wireless Networks 27

28 Wireless channel: empirical models More complex scenarios can hardly be modeled with closed formulas Empirical models are usually adopted where received power is d -η P R = P T g T g R λ 4π where the η is the propagation factor which typically ranges between 2 and 5 More complex empirical models (e.g. Hata) take into account several parameters including scenario (urban, rural), antenna heights, etc. 2 d η Antonio Capone: Wireless Networks 28

29 Wireless channel: empirical models Okumura/Hata formula: LP = log f 3.82 log ht a( hr ) + + ( log ht ) log d [db] where f frequency in MHz (from 5 to 5 MHz) h T base station height (in m) h R mobile station height (in m) a(h R ) correlation factor depending on area shape d distance (in km) For a 9 MHz system, h T = 3 m, a(h R ) : L P = log d Antonio Capone: Wireless Networks 29

30 Wireless channel: multipath fading Copies of the same signal arrive from different paths Their combination at the receiver depends on: Number of copies Relative shift amplitude frequency Antonio Capone: Wireless Networks 3

31 Wireless channel: multipath fading,5,5 -, s(t) s(t+t) s(t)+s(t+t) The resulting signal can be attenuated - -,5 T=4/5π 2,5 2,5 s(t) s(t+t) s(t)+s(t+t) or amplified,5 -, ,5-2 T= π /6-2,5 Antonio Capone: Wireless Networks 3

32 Wireless channel: shadowing Signal can be partially absorbed or reflected by obstacles Further attenuation called shadowing Antonio Capone: Wireless Networks 32

33 Mobility management In wireless networks, users can roam in the service area moving among cells This require an adaptive routing of information based on user position in the network All wireless networks have a set of mobility management mechanisms to track user position Antonio Capone: Wireless Networks 33

34 Mobility management: cellular systems In cellular systems mobility management adopts different procedures based on user state IDLE (no active call) or ACTIVE (in conversation) ACTIVE: dynamic call rerouting cell-bycell (Handover) IDLE: user position tracking (Cell selection, Location Update, Paging) Antonio Capone: Wireless Networks 34

35 Mobility management: Cell selection Base stations transmit periodically general system information and cell identity on a broadcast channel User terminals scan all channels to received broadcast transmissions from nearby base stations User terminal select autonomously the best cell, usually based on signal strength Antonio Capone: Wireless Networks 35

36 Mobility management: Location Update Location Area: set of cells User position tracking is based on location areas and not on cells The currently visited LA is stored in a data base Data Base LA LA 2 Antonio Capone: Wireless Networks 36

37 Mobility management: Location Update When a mobile terminal in idle state move in a different LA a Location Update procedure is started Information in the data base is used to route incoming call to mobile terminals LA Data Base LA 2 Antonio Capone: Wireless Networks 37

38 Mobility management: Paging When a call must be routed to a mobile terminal the currently visited LA is retrieved from the data base Then the paging procedure is started All base stations in the LA broadcast a paging message with the ID of the called user When the mobile terminal replies the call is routed paging paging reply Data Base Antonio Capone: Wireless Networks 38

39 Mobility management: Handover Handover is always triggered by the network based on measurements performed by the mobile terminal (received powers, quality, etc.) Handover procedures must be fast to avoid quality degradation h Handover TH Receiver TH Δt t Antonio Capone: Wireless Networks 39

40 Mobility management: Handover When starting a handover? Due to signal fluctuations call may be switched back to original cell (ping-pong effect) Antonio Capone: Wireless Networks 4

41 Mobility management: Handover Hard Handover (GSM-2G) Soft Handover (UMTS-3G) Antonio Capone: Wireless Networks 4

42 Mobility management: Data networks (WLAN, WMAN, ) Within the same network, mobility is managed at layer two Among different networks mobility is managed at layer three (using e.g. Mobile IP) AP AP2 R A A Antonio Capone: Wireless Networks 42

43 Part A Multiple access Antonio Capone: Wireless Networks 43

44 Multiplexing and multiple access Different information flows sharing the same physical channel One transmitting station: multiplexing Many transmitting stations (one per flow): multiple access Antonio Capone: Wireless Networks 44

45 Multiplexing and multiple access Node Node 2 Multiplexing MPX DMPX Multiple Access Node Node 2 Node 3 Node 4 AM AM AM AM Broadcast channel Antonio Capone: Wireless Networks 45

46 Wireless networks: Multiplexing One transmitting station channel channel 3 channel 2 Typical problem in in downlink (forward link) of of cellular systems (base station mobile users) Antonio Capone: Wireless Networks 46

47 Wireless networks: Multiple access Several transmitting stations (coordination problem) channel channel 2 channel 3 Typical uplink (reverse link) problem in in cellular systems (from users to to base station) A duplexing technique is also needed for sharing between uplink and downlink channels Antonio Capone: Wireless Networks 47

48 Wireless networks: Frequency reuse The radio resource is limited and can not be exclusively dedicated to a channel in a cell The same radio resource is used in different cells sufficiently far apart to not interfere Critical problem with a trade off between number and quality of channels see later on Antonio Capone: Wireless Networks 48

49 Multiple access From now on Multiple access includes also multiplexing and duplexing Multiple access at the physical layer: A single channel is divided into subchannels using physical parameters (frequency, time, code) static resource management Multiple access at logical layer: packet access with logical information in the packet header and distributed coordination mechanisms dynamic resource management In real systems different multiple access techniques at physical and logical layers are usually combined together Antonio Capone: Wireless Networks 49

50 FDM/FDMA (Frequency Division Multiplexing/Multiple Access) Available bandwidth is divided into subbands and assigned to different subchannels Simple technique used basically in all systems f mod. f min f max Antonio Capone: Wireless Networks 5

51 TDM/TDMA (Time Division Multiplexing/Multiple Access) Time is divided into slots Groups of N consecutive slots are organized into frames A subchannel can use a given slot in all frames slot frame frame Antonio Capone: Wireless Networks 5

52 TDMA: Guard time Tg = max(2 i τ i ) Antonio Capone: Wireless Networks 52

53 TDMA: reduced guard time Timing Advance: If the propagation delay τ is known it can be compensated anticipating the transmission (centralized access only!) τ must be dynamically estimated and signaled back to the mobile 2) Delay estimation 3) Delay signaled back 4) Subsequent transm. with reduced guard time ) First transm. Antonio Capone: Wireless Networks 53

54 CDM/CDMA (Code Division Multiplexing/Multiple Access) Symbols (bits) on the channels are multiplied by a code In CDM codes are orthogonal, while in CDMA they have limited correlation C (t) T C ( t) C 2 ( t) = C 2 (t) N i= c i c 2i = Antonio Capone: Wireless Networks 54

55 CDM/CDMA (Code Division Multiplexing/Multiple Access) s s s 2 C C C + 2 C 2 s 2 s N s N C N N sici Ck = T i= s k C N Antonio Capone: Wireless Networks 55

56 CDMA: spreading and despreading The code expands the radio bandwidth of the signal S(f) B f Spreading of the radio spectrum S M (f) nb f n number of chips in the code: spreading factor (SF) Antonio Capone: Wireless Networks 56

57 CDMA: spreading and despreading Different signals use the same radio band s M (t) s M2 (t) + nb f Antonio Capone: Wireless Networks 57

58 CDMA: spreading and despreading At receiver the signal is multiplied by the code (de-spreading) nb f De-spreading B f The interfence of the other signals is reduced by /n B f Antonio Capone: Wireless Networks 58

59 Packet access At logical layer multiple access can be managed in a dynamic and distributed way using multiple access protocols First multiple access protocols have been designed for LANs Nowadays multiple access protocols are mainly used in wireless networks (no more shared medium wired LANs) Antonio Capone: Wireless Networks 59

60 Packet access: Classification Scheduled access Transmissions on the channel are sequential with no conflicts Polling schemes Centralized scheduling schemes Random access Transmission are partially uncoordinated and can overlap (collision) Conflicts are resolved using distributed procedures based on random retransmission delay Antonio Capone: Wireless Networks 6

61 Packet access: Local and global queues In the case of multiplexing (single station) we have a single queue that is managed according to a scheduling algorithm In case of multiple access with M stations with local queues we still have the opportunity to use a single virtual queue at a central decision point that schedule access to the channel... Antonio Capone: Wireless Networks 6

62 Packet access: Local and global queues However, to inform the scheduler of the status of the local queues and provide access grants we have to use the channel (coordination signaling) The centralized scheduling approach is quite flexible but complex It is adopted in several wireless technologies like e.g. WiMax Antonio Capone: Wireless Networks 62

63 Assumptions and notation In the following we drop the assumption of global coordination and analyze distributed mechanisms Let us assume that arrival times in the M local queues are described by a Poisson process with rate λ/m (λ global rate) The system status is described by vector n = ( n, ) n2,..., n M Where n i is the number of packets in queue i The system evolution is described by the process N(t) Antonio Capone: Wireless Networks 63

64 Polling Polling schemes are scheduled access schemes where stations access the channel according to a cyclic order The polling message, or token, is the grant for access the channel The token can be distributed by a central station (roll-call polling) or passed from station to station (hub polling or token system) Let us assume that packet transmission time is T and that token passing time is h, both constant Polling schemes differentiate based on the service policy (exhaustive, gated, limited) Antonio Capone: Wireless Networks 64

65 Exhaustive Polling With exhaustive polling, stations when receive the token transmit all packets in the queue before releasing it Let us analyze the behavior of this system The probability that the channel is transmitting a packet at a random time t is give by ρ = λt Antonio Capone: Wireless Networks 65

66 Exhaustive Polling The average waiting time E[W] in the queue can be calculated considering three components E [ W ] = W + W + W 2 3 Arrival in queue 8 transmission Antonio Capone: Wireless Networks 66

67 Exhaustive Polling W E[ Nc ] T = E[Nc] is the average number of packets transmitted before considered packet Using Little s result is can be expressed as: E[ N c ] = λe[ W ] λ a T Therefore: W = λ TE[ W ] = ρe[ W ] N Antonio Capone: Wireless Networks 67

68 Exhaustive Polling W W 2 3 T = ρ + ( ρ) 2 M = h 2 h 2 The total average waiting time is given by: E[ W ] T = ρe[ W ] + ρ + ( ρ) 2 h 2 + ( M ) 2 h Antonio Capone: Wireless Networks 68

69 Exhaustive Polling Solving by E[W] we get: E[ W ] = ρ 2( ρ) T + M 2( ρ ρ) h Waiting time of a single queue (M/D/) Additional waiting time due to token passing time Note that: ρ max = Antonio Capone: Wireless Networks 69

70 Exhaustive Polling The average token cycle time is given by the transmission time of all packets that arrive during a cycle plus the token passing time [ ] = λ [ ] E C E C T + Mh [ ] E C = Mh ρ Antonio Capone: Wireless Networks 7

71 Gated Polling With gated polling, stations when receive the token can transmit all packets that are in queue at the time when the token arrives The expression of the average waiting time is similar to previous case with an additional term This is the additional cycle the packet has to wait when it arrives when the token is already at the station ρ W 4 = Mh = M hρ Antonio Capone: Wireless Networks 7

72 Gated Polling Therefore we get: E[ W ] = ρ 2( ρ) T + M 2( + ρ ρ) h Again ρ max = Antonio Capone: Wireless Networks 72

73 Limited Polling With limited polling, stations when receive the token can transmit only up to k packets The special case of k= is called Round-Robin Here we have one more additional term which are the additional cycles the packet has to wait, one per each packet in the queue at the arrival moment E[ Nc ] W5 = Mh = λe[ W ] h M Antonio Capone: Wireless Networks 73

74 Limited Polling Therefore we get: E[ W ] = 2( ρ h ρ + T T ) T + M + ρ h + T 2( ρ T ) h Now we have: ρ max = T T + h Antonio Capone: Wireless Networks 74

75 Polling in real networks There are several examples where polling is used for regulating access to a channel in wireless technologies WiFi (Point Coordination Function PCF or HCF Hybrid Coordination Function) Bluetooth The main difference with simple schemes we considered so far is that the station sequence can be dynamically changed Antonio Capone: Wireless Networks 75

76 Polling in Bluetooth Master S S S SB Slave Slave 2 P S SB S M S P P S SB Slave 3 SCO (Synchronous Connection Oriented) ACL (Asynchronous ConnectionLess) Antonio Capone: Wireless Networks 76

77 Polling in WiFi (IEEE 82.e) Antonio Capone: Wireless Networks 77

78 Random access With random access there are possible conflics on the channel (collisions) Conflicts are resolved using the channel feedback and some procedure to select a random waiting time The minimum channel feedback a station need to have is the information if its transmission was sccessful or not The first and simplest random access protocol is Aloha which uses just this minimum feedback Antonio Capone: Wireless Networks 78

79 AlohaNet The ALOHA network was created at the University of Hawaii in 97 under the leadership of Prof. Norman Abramson It was the first wireless network! Antonio Capone: Wireless Networks 79

80 Aloha The access mechanism is very simple: When there is a packet to be transmitted, just transmit it. If transmission fails, wait for a random time and retransmit T T Antonio Capone: Wireless Networks 8

81 Aloha Let us assume the transmission starting times on channel are a Poisson process with rate λ Let us consider the normalized rate G=λT The success probability is given by the probability that there is no other transmission in a 2T interval P = e s 2G The normalized throughput S is therefore given by: S = Ge 2G Antonio Capone: Wireless Networks 8

82 Aloha If transmissions are somehow synchronized (slotted Aloha) the vulnerability period reduces to T and therefore S = Ge G Infinite population model Antonio Capone: Wireless Networks 82

83 Aloha: Single buffer Unfortunately, the traffic on the channel is the combination of new transmissions and retransmissions and it can increase if throughput reduces To evaluate the dynamic behavior of Aloha let us consider an enhanced model Let us assume we have M stations with a single transmission buffer (max packet) Channel is slotted If buffer is empty a new packet arrive and is immediately transmitted in the slot with probability α If buffer is full, packet is retransmitted with probability β Antonio Capone: Wireless Networks 83

84 Aloha: Single buffer The system status is given by n(t), the number of full buffers at time t n(t) is a discrete Markov chain The probability that in a slot there are i arrivals given n buffers are full is: M n a( i, n) α α i i M n i = ( ) While the probability that there are i retransmission is: n b( i, n) β β i i n i = ( ) Antonio Capone: Wireless Networks 84

85 Aloha: Single buffer a(,n)b(,n)+ +a(,n)[-b(,n)] a(,n)[-b(,n)] a(n,i) n- n n+... n+i a(,n)b(,n) Antonio Capone: Wireless Networks 85

86 Aloha: Single buffer We can solve (numerically) the chain and get the stationary state probability π n The throughput S is given by S = E[ s( n)] = π ns( n) s( n) M n= = a(, n) b(, n) + a(, n) b(, n) Antonio Capone: Wireless Networks 86

87 Aloha: Single buffer The traffic on the channel is: g ( n) = ( M n) α + nβ and the new arrivals: a( n) = ( M n)α We can express n as a function of g: And calculate n = g Mα β α a( g) = Mα ( g Mα) α β α Antonio Capone: Wireless Networks 87

88 Aloha: Single buffer Similarly, we can also get s(g) using stationary probabilities (the curve is very close to that of the infinite population model) We can consider points where s(g)=a(g) as equilibrium points of the process Equilibrium points can be stable or unstable Antonio Capone: Wireless Networks 88

89 Aloha: Single buffer Antonio Capone: Wireless Networks 89

90 Aloha: Single buffer Antonio Capone: Wireless Networks 9

91 Aloha: Single buffer Antonio Capone: Wireless Networks 9

92 Aloha in real networks There are several technologies where Aloha is still adopted including: Random access signaling channel of cellular systems (like e.g. GSM) Reservation channel of WiMax RFID Antonio Capone: Wireless Networks 92

93 Carrier Sense Multiple Access (CSMA) If the channel feedback is richer, more efficient random access mechanisms can be adopted If the propagation time is short wrt to transmission time we can sense the channel status CSMA: like Aloha but transmit only when you sense the channel free a=τ/t τ τ Antonio Capone: Wireless Networks 93

94 Carrier Sense Multiple Access (CSMA) On the channel we have cycles of Busy (at least one station sense the channel as busy) and Idle (all stations sense the channel free) periods The throughput S can be given by: S = α B + I where B and I are the average busy and idle periods and α is the probability that there is a success transmission in a busy period Antonio Capone: Wireless Networks 94

95 Carrier Sense Multiple Access (CSMA) Making the same assumptions of the aloha infinite population model we have: ag α = e I = G B = e ag ( + a) + ( e ag where Z is the time when colliding transmission partially overlap )( + a + Z) Antonio Capone: Wireless Networks 95

96 Carrier Sense Multiple Access (CSMA) It can be shown that: Z ae = a + e Therefore we get: ag ag G S = ag Ge G( + 2a) + e ag Antonio Capone: Wireless Networks 96

97 Carrier Sense Multiple Access (CSMA) Antonio Capone: Wireless Networks 97

98 CSMA in real networks There are several technologies that are based on variants of the CSMA protocol Including Ethernet Today the most famous and widely used one is WiFi Antonio Capone: Wireless Networks 98

99 IEEE 82. random access source DIFS RTS DATA SIFS SIFS SIFS destination CTS ACK neighbors NAV (RTS) NAV (CTS) Random Backoff Antonio Capone: Wireless Networks 99

100 IEEE 82. random access Similarly to the general model we can derive a model for WiFi α = e I = G ag a = interframe space b = duration of RTS and CTS B = e ag ( + 3a + 2b) + ( e ag )( b + a + Z) S = G( + 2a) + e ag ag Ge G( b)( e ag ) + (2a + 2b) Ge ag Antonio Capone: Wireless Networks

101 Part B Resource and radio planning Antonio Capone: Wireless Networks

102 Summary Introduction to radio planning Coverage planning Capacity planning (frequency assignment) Joint coverage and capacity planning (planning of CDMA systems) References: [] E. Amaldi, A. Capone, F. Malucelli, C. Mannino, Optimization Problems and Models for Planning Cellular Networks, in Handbook of Optimization in Telecommunications, Ed. P.M. Pardalos and M.G.C. Resende, Kluver Academic Publishers, 25 (available at [2] IEEE Wireless Communications Magazine, special issue on 3G/4G/WLAN/WMAN radio planning and optimization, Eds. A. Capone and J. Zhang, to appear December 26. Antonio Capone: Wireless Networks 2

103 Resource planning One we divide the available spectrum into sub-channels using some multiple access technique we need to plan how sub-channels are used in the cells (these includes also data networks like wifi and wimax) This problem is strictly related to the general problem of designing the radio access part of the network (radio planning) Antonio Capone: Wireless Networks 3

104 What is radio planning? When we have to install a new wireless network or extend an existing one into a new area, we need to design the fixed and the radio parts of the network. This last phase is called radio planning. Antonio Capone: Wireless Networks 4

105 What is radio planning? The basic decisions that must be taken during the radio planning phase are: Where to install base stations (or access points, depending on the technology) How to configure base stations (antenna type, height, sectors orientation, tilt, maximum power, device capacity, etc.) X X X X Antonio Capone: Wireless Networks 5

106 Radio Planning When planning and optimizing a cellular system, a number of aspects must be considered, including signal propagation, traffic estimation, antenna positioning, antenna configuration, interference. Here we ll focus on the decision problems that give rise to interesting and challenging mathematical programming models which must account for the peculiarities of the specific network technology. Antonio Capone: Wireless Networks 6

107 Propagation prediction One of the key elements for the radio planning is propagation prediction that allows to estimate the area covered by each base station The covered area is the area where the received signal strength is R above a threshold Received signal strength depends on emitted power and path loss Antonio Capone: Wireless Networks 7

108 Propagation prediction (2) Path Loss depends on several phisical effects related to the propagation of electromagnetic waves including: Distance Frequency Ground morfology Atmosferic phenomena Antenna heights Etc. Path loss has been modeled using many propagation models (Okumura-Hata, Cost23, Walfish-Ikegami, Bertoni, etc.) that can be grouped into three categories: Empirical Statistical Deterministic Antonio Capone: Wireless Networks 8

109 Propagation prediction (3) Due to the complex propagation environment of cellular systems simple statististical models are aften adopted Deterministic techniques (e.g. Ray tracing) are sometime used for indoor propagation A deterministic component due to the distance is the starting point of statistical models Path loss: L p ( d) = A+ Blog( d) [db] Constants are estimated using empirical models (e.g. Okumura-Hata) or measurements Antonio Capone: Wireless Networks 9

110 Propagation prediction (4) If necessary two or three linear models are combined: SS [dbm] Measured Signal Strength Dual-slope prediction -slope prediction Survey Meters Statistical dispersion of data is taken into account modeling shadow fading and multipath fading using random variables. Antonio Capone: Wireless Networks

111 Traffic estimation Traffic distribution in the service area is usually hard to predict in the radio planning phase since it depends on several issues including area population, buildings, market penetration of the considered service, etc. Traffic distribution is usually provided using a discrete set of points I, test points (TP), that are considered as centroids of traffic Antonio Capone: Wireless Networks

112 Antenna positioning The selection of possible antenna sites depends on several technical (traffic density and distribution, ground morphology, etc.) and non-technical (electromagnetic pollution, local authority rules, agreements with building owners, etc.) issues. We denote with S the set of candidate sites (CS) We can assume that the channel gain g ij between TP i and CS j is provided by a propagation prediction tool Antonio Capone: Wireless Networks 2

113 Antenna configuration Radiation diagram Horizontal (sectors) and vertical (tilt) angles Maximum emission power (pilot channel power) Height Base station capacity Etc. Antonio Capone: Wireless Networks 3

114 Antenna configuration (2) The antenna configuration affects only the signal level received at TPs For each CS j we can define a set of possible antenna configurations K j We can assume that the channel gain g ijk between TP i and CS j depends also on configuration k. Based on signal quality requirement and channel gain we can evaluate if TP i can be covered by CS j with an antenna with configuration k, And define coefficients: a ijk = if TP i can be covered by CS j with conf. k otherwise Antonio Capone: Wireless Networks 4

115 S Summarizing {,..., m} for each I a = = ijk {,..., n} set of j S, K set of j candidatesites(cs) set of configurat ions test points(tp) if TPi is coveredby CS j with conf. k = otherwise i I, j S, k K j Antonio Capone: Wireless Networks 5

116 Interference Multiple access techniques are used to define communication channels on the available radio spectrum FDMA TDMA CDMA TIME FREQUEN CY POW ER TIME FREQUENC Y POW ER TIME FREQUENC Y POW ER Radio resources for wireless systems are limited and must be reused in different areas (cells) Resource reuse generates interference Antonio Capone: Wireless Networks 6

117 Interference (2) Interference can be tolerated (good communication quality) if the Signal-to-Interference Ratio (SIR) is high enough SIR constraint limits the number of simultaneous communications per cells, i.e. the system capacity Capacity is another key element that must be considered during radio planning FDMA/TDMA cellular systems adopt a two phases radio planning Coverage planning Capacity planning (frequency assignment) CDMA cellular systems require a single phase approach Joint coverage and capacity planning Antonio Capone: Wireless Networks 7

118 Coverage planning Antonio Capone: Wireless Networks 8

119 Coverage planning The goal of the coverage planning phase is to: Select where to install base stations Select antenna configurations In order to guarantee that the signal level in all TPs is high enough to guarantee a good communication quality Note that interference is not considered in this phase Antonio Capone: Wireless Networks 9

120 Decision variables and parameters Decision variables: if a basestation with configuration k is installedin CS j = otherwise Installation c costs: jk y jk Cost related to the installation of a base station in CS j with configuration k Antonio Capone: Wireless Networks 2

121 Set covering problem (SCP) min j S j S k K y jk k K k K j y j j jk c a jk ijk y y jk jk j S i I {, } j S, k K j Objective function: total network cost Full coverage constraints One configuration per site Integrality constraints P { i a =} = ijk Let jk Variables y jk define a subset Such that P I U* j S jk = S * S Antonio Capone: Wireless Networks 2

122 Set covering problem (SCP) SCP is NP-hard However several efficient algorithms has been proposed (see [3] for a survey) Even simple greedy algorithms allow to obtain high quality solutions [3] S. Ceria, P. Nobili, and A. Sassano. Set covering problem. In M. Dell Amico, F. Maffioli, and S. Martello, editors, Annotated Bibliographies in Combinatorial Optimization, chapter 23, pages John Wiley and Sons, 997. Antonio Capone: Wireless Networks 22

123 Greedy algorithm for SCP Step set S*= Step if P j = for all j then STOP Otherwise find k (J-J*) such that: is maximum cj Step 2 S*:=S* {k} P j :=P j -P k j Go to Step. Note: we don t consider configurations here for the sake of simplicity P j Antonio Capone: Wireless Networks 23

124 Antonio Capone: Wireless Networks 24 Greedy algorithm: Example () Step : S * = = Π = = C V

125 Antonio Capone: Wireless Networks 25 Greedy algorithm: Example (2) Step : k=5 = Π = = C V

126 Antonio Capone: Wireless Networks 26 Greedy algorithm: Example (3) Step 2: S * ={5},... = Π = = C V

127 Antonio Capone: Wireless Networks 27 Greedy algorithm: Example (4) Step 2: ricalculate V e Π Π = = V

128 Antonio Capone: Wireless Networks 28 Greedy algorithm: Example (5) Step : k= Step 2: S * ={5,}, ricalculate V e Π Π = = V

129 Antonio Capone: Wireless Networks 29 Greedy algorithm: Example (6) ricalculate V e Π Π = = 2 V

130 Antonio Capone: Wireless Networks 3 Greedy algorithm: Example (7) Step : k=2 Step 2: J * ={5,,2}, ricalculate V e Π Π = = 2 V

131 Antonio Capone: Wireless Networks 3 Greedy algorithm: Example (8) ricalculate V e Π Π = = V

132 Antonio Capone: Wireless Networks 32 Greedy algorithm: Example (9) Step : k=3 Step 2: J * ={5,,2,3}, ricalculate V e Π Π = = V

133 Antonio Capone: Wireless Networks 33 Greedy algorithm: Example () ricalculate V e Π STOP Π = = V

134 Antonio Capone: Wireless Networks 34 Greedy algorithm: Example () In this simple example it s easy to observe that the solution J * = {5,,2,3} is sub-optimal In fact this solution has a lower cost: = Π = = C V

135 Maximum coverage problem (MCP) In practice the coverage requirement is often a soft constraints and the problem actually involves a tradeoff between coverage and installation costs max λ j S k K y z i i I jk y z i k K j j jk a ijk j S {, } y k K jk c z j S {, } i I j i i I j S, k K Antonio Capone: Wireless Networks 35 jk y jk Objective function: trade-off between cost and coverage j Definition of variables z One configuration per site Integrality constraints

136 Assigning test points to base stations When a TP is covered by more than one base station: j S k K j a ijk y jk # of base stations covering TP i the serving base station is not defined We can define new assignment variables: x ij if TPi isassignedtocsj = otherwise Antonio Capone: Wireless Networks 36

137 Set covering with assignment (SCA) min j S j S k K x y x ij ij jk k K x j ij y j jk c k K jk = j a y {, } i I ijk jk j S y jk Definition of variables x j S, k K {, } i I, j S Coverage constraints j Antonio Capone: Wireless Networks 37

138 Capacity constraints Obviously, without additional constraints SCA provides the same solution as SCP Using x variables we can add constraints on cell capacity: i I d x i ij v k K j jk j S where d i is the traffic demand associate to TP i and v jk is the capacity of a base station in CS j with configuration k Other constraints related to cell shape can be added y jk Antonio Capone: Wireless Networks 38

139 Assignment to the nearest base station One of these rules is the requirement of assigning a TP to the closest (in terms of signal strength) activated BS. One way to express this constraint for a given TP i is to consider all the pairs of BSs and configurations that would allow connection with i and sort them in decreasing order of signal strength. Let {( j }, k),( j2, k2),..., ( j L, kl) be the ordered set of BS-configuration pairs The constraints enforcing the assignment on the nearest BS are: y j l k l + L h= l+ x ij h l L Antonio Capone: Wireless Networks 39

140 Capacity planning (frequency assignment) Antonio Capone: Wireless Networks 4

141 Cluster model After coverage planning, capacity planning is in charge of defining which radio resources can be used by each cell The amount of resources (frequencies) assigned to cells determines system capacity Frequencies can be reused, but SIR (quality) constraints must be enforced A simple didactical model considers hexagonal cells and homogeneous traffic Frequencies are divided into K groups and assigned to a group of K cells, named cluster. The cluster is repeated in the area in a regular fashion F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 F 2 F 7 F 3 F F 6 F5 F 4 Antonio Capone: Wireless Networks 4

142 Cluster model (2) Only some values of K are admissible K=,3,4,7,9,2,3, Given the minimum value of SIR, SIR min, we can determine K the minimum value of K Received power: P r = P G d t η Antonio Capone: Wireless Networks 42

143 Antonio Capone: Wireless Networks 43 Cluster model (2) d r D d d 2 d 3 d 4 d 5 d6 Same antennas and same power: = = = = = 6 6 i i i i t t d d d G P d G P SIR η η η η Worst case d = r Approxmation d i = D η η η = R D r SIR 6 6

144 Cluster model (3) SIR depends only on the reuse ratio R=D/r and not on the power and the cell radius Geometric consideration provides: K = 2 R 3 And therefore: K = min ( 6SIR) 3 2/ η Antonio Capone: Wireless Networks 44

145 Graph based models Unfortunately cells are not hexagonal and traffic is not homogeneous Other models have been proposed for practical cases (see [6] for a quite complete survey) Some popular models are based on graph coloring problems [6] K. Aardal, S.P.M. van Hoesel, A. Koster, C. Mannino, and A. Sassano. Models and solution techniques for frequency assignment problems. 4OR, (4):26 37, 23. Antonio Capone: Wireless Networks 45

146 Graph based models (2) Compatibility graph G(V,E) Vertices are base stations Two vertices are connected by an edge if the two base stations cannot reuse the same frequencies Antonio Capone: Wireless Networks 46

147 Graph based models (3) Any coloring of the vertices of G (i.e., assignment of colors such that adjacent vertices have different colors) is an assignment of frequencies to the network such that no mutual interfering BSs receive the same frequency. A minimum cardinality coloring of G is a minimum cardinality non-interfering frequency assignment of the network. Graph coloring problem is NP-hard and several exact algorithms and heuristics have been proposed. This simple model assumes: One frequency per BS Two distinct frequencies do not interfere Antonio Capone: Wireless Networks 47

148 Graph based models (4) Generalized graph coloring models: Compatibility matrix: Frequencies are numbered according spectrum position Sets F j defines assignments of frequencies to BSs F Traffic constraints: j Compatibility constraints: C= { c ij } i, j S = m j j S f i f c i, j S, f F, j ij i i f j F j [7] W.K. Hale. Frequency assignment: Theory and applications. Proceedings of the IEEE, 68:497 54, 98. Antonio Capone: Wireless Networks 48

149 Graph based models (5) if c ij = i and j can reuse the same frequencies if c ij = i and j cannot use the same frequencies if c ij =2 i and j cannot use either the same frequencies and adjacent frequencies i Optimization objective: c j ij Min Span(G) = number of frequencies used MS-FAP (Minimum Span Frequency Assignment Problem) Antonio Capone: Wireless Networks 49

150 Graph based models (6) Comments: Graph based models do not consider SIR constrains explicitly The cumulative effect of interference is not accouter for Compatible? Antonio Capone: Wireless Networks 5

151 Interference based models Minimum Interference Frequency Assignment Problem (MI-FAP) Generalization of the max k-cut problem on edge-weighted graphs penalty p vwfg v, w S, f, g F representing the interference (cost) generated when v is assigned with f and w is assigned with g. Decision variables: x vf = if f isassignedtov otherwise Antonio Capone: Wireless Networks 5

152 Interference based models (2) Objective function: min p v, w S f, g F vwfg x vf x wg The problem can be linearized: x z vwfg vf if f isassignedtovandg = otherwise + x + z v, w S f, wg vwfg isassignedtow g F Antonio Capone: Wireless Networks 52

153 Interference based models (3) Linear MI-FAP: min v, w S f, g F p vwfg z vwfg s.t. x vf + x + z v, w S f, wg vwfg g F f F x vf = m( v) v S Antonio Capone: Wireless Networks 53

154 Interference based models (4) MI-FAP: Total interference minimized No control on single interference values MI-FAP variants account for explicit SIR constraints (see [8,9]) [8] M. Fischetti, C. Lepschy, G. Minerva, G. Romanin-Jacur, and E. Toto. Frequency assignment in mobile radio systems using branch-and-cut techniques. European Journal of Operational Research, 23:24 255, 2. [9] A. Capone and M. Trubian. Channel assignment problem in cellular systems: A new model and tabu search algorithm. IEEE Trans. on Vehicular Technology, 48(4): , 999. Antonio Capone: Wireless Networks 54

155 Politecnico di Milano Advanced Network Technologies Laboratory Joint coverage and capacity planning

156 Approaches to the radio planning 2nd Generation Systems (GSM, D-AMPS, D...) two-phases approach ) Radio coverage minimum signal level in all the service area 2) Frequency assignment meet traffic constraints meet quality (SIR) constraints 3rd Generation Systems based on W-CDMAW two-phases approach not suitable because: no frequency planning for CDMA power control determines the cell breathing effect Planning must also consider traffic demand distribution SIR constraints Antonio Capone: Wireless Networks 56

157 Covering traffic in W- CDMA systems Traffic generated can be considered covered (served) by the system if the QUALITY of the connection is good Quality measure: Signal-to- Interference Ratio (SIR) SIR SIR downlink uplink = = SF SF α I + I +η I Antonio Capone: Wireless Networks 57 out out P P + I rec rec in in +η

158 Power Control (PC) mechanism Dynamic adjustment of the transmitted power to minimize interference Two PC mechanisms: Power-based PC emission powers are adjusted so that received powers are equal to a given P tar SIR-based PC emission powers are adjusted so that all SIR are equal to a given estimated SIR tar Antonio Capone: Wireless Networks 58

159 Cell breathing effect Due to the power limitations the area actually covered by a BS depends on interference (traffic) level When traffic (interference) increases the SIR constraint cannot be met for terminals far from the BS due to higher channel attenuation Since only terminals close to the BS can be actually served it is as if the actual cell area reduces Since this phenomenon affects coverage, traffic levels must be carefully considered during radio planning Antonio Capone: Wireless Networks 59

160 Joint coverage and capacity planning set of candidate sites where to install BSs: S={,,m} installation costs: c j, j S set of test points (TPs): I={,,n} traffic demand: a i, i I equivalent users: u i =φ(a i ) propagation gain matrix: G=[g ij ], i I, j S Problem: Select a subset of candidate sites where to install BSs, and assign TPs to BSs so that quality constraints are satisfied and the total cost is minimized Antonio Capone: Wireless Networks 6

161 Joint coverage and capacity planning (2) Decision variables: y x i ij = = if Basic constraints: j S x x ij ij x y,y ij j a BSis installed in j S otherwise if test point i I is assigned to BS j S otherwise i I assignment j i I, j S {, } i I, j S coherence integrality Antonio Capone: Wireless Networks 6

162 Joint coverage and capacity planning (3) Objective function: max i I j S u i x ij λ j S c j y j maximize minimize covered traffic installation costs We assume a power-based Power Control (received power = P tar ) variables x ij are defined only for pairs such that: tar power limit P g ij P max Antonio Capone: Wireless Networks 62

163 Joint coverage and capacity planning (4) SIR constraints: h I u h g hj t S Ptar P g tar ht total interference x SIR bilinear constraints which can be easily linearized: P tar with a value of M large enough ht + M j h I t S signal power min y j j ( ) hj y SIR u x j S min h g g ht ht S Antonio Capone: Wireless Networks 63

164 Joint coverage and capacity planning (5) Solution approach: State-of-the-art ILP solvers can provide the exact solution only for very small instances Heuristics have been proposed Promising approach based on Tabu Search [] E. Amaldi, A. Capone, and F. Malucelli. Planning UMTS base station location: Optimization models with power control and algorithms. IEEE Transactions on Wireless Communications, 2(5): , 23. [] E. Amaldi, P. Belotti, A. Capone, F. Malucelli, Optimizing base station location and configuration in UMTS networks, Annals of Operations Research, vol. 43, June 26. Antonio Capone: Wireless Networks 64

165 Politecnico di Milano Advanced Network Technologies Laboratory Thank you! Antonio Capone Politecnico di Milano Advanced Network Technologies Laboratory Contact: Web: (personal) (laboratory)