EE360: Lecture 7 Outline Cellular System Capacity and ASE Announcements Summary due next week

Transcription

1 EE360: Lecture 7 Outlne Cellular System Capacty and ASE Announcements Summary due next week Capacty Area Spectral Effcency Dynamc Resource Allocaton

2 Revew of Cellular Lecture Desgn consderatons: Spectral sharng, reuse, cell sze Evoluton: 1G to 2G to 3G to 4G and beyond Multuser Detecton n cellular MIMO n Cellular Multuser MIMO/OFDM Multplexng/dversty/IC tradeoffs Dstrbuted antenna systems Vrtual MIMO 8C Cmn-7/98

3 Cellular System Capacty Shannon Capacty Shannon capacty does no ncorporate reuse dstance. Wyner capacty: capacty of a TDMA systems wth jont base staton processng User Capacty Calculates how many users can be supported for a gven performance specfcaton. Results hghly dependent on traffc, voce actvty, and propagaton models. Can be mproved through nterference reducton technques. Area Spectral Effcency Capacty per unt area In practce, all technques have roughly the same capacty for voce, but flexblty of OFDM/MIMO supports more heterogeneous users

4 Defnng Cellular Capacty Shannon-theoretc defnton Multuser channels typcally assume user coordnaton and jont encodng/decodng strateges Can an optmal codng strategy be found, or should one be assumed (.e. TD,FD, or CD)? What base staton(s) should users talk to? What assumptons should be made about base staton coordnaton? Should frequency reuse be fxed or optmzed? Is capacty defned by uplnk or downlnk? Capacty becomes very dependent on propagaton model Practcal capacty defntons (rates or users) Typcally assume a fxed set of system parameters Assumptons dffer for dfferent systems: comparson hard Does not provde a performance upper bound

5 Approaches to Date Shannon Capacty TDMA systems wth jont base staton processng Multcell Capacty Rate regon per unt area per cell Achevable rates determned va Shannon-theoretc analyss or for practcal schemes/constrants Area spectral effcency s sum of rates per cell User Capacty Calculates how many users can be supported for a gven performance specfcaton. Results hghly dependent on traffc, voce actvty, and propagaton models. Can be mproved through nterference reducton technques. (Glhousen et. al.)

6 Wyner Uplnk Capacty Lnear or hexagonal cells Receved sgnal at base staton (N total users) Propagaton for out-of-cell nterference captured by a Average power constrant: E Capacty C N defned as largest achevable rate (N users)

7 Lnear Array Theorem: lm N C N C*( a) for Optmal scheme uses TDMA wthn a cell - Users transmt n 1/K tmeslots; power KP Treats co-channel sgnals as nterference:

8 Results Alternate TDMA CDMA w/ MMSE

9 Channel Reuse n Cellular Systems Channel Reuse n Cellular Systems Motvaton: power falloff wth transmsson dstance Pro: ncrease system spectral effcency Con: co-channel nterference (CCI) Channel : tme slot, frequency band, (sem)-orthogonal code... Cellular Systems wth dfferent multple-access technques CDMA (IS-95, CDMA2000): weak CCI, channel reuse n every cell codes desgned wth a sngle and narrow autocorrelaton peak TDMA (GSM), FDMA (AMPS): much stronger CCI a mnmum reuse dstance requred to support target SINR Channel reuse: tradtonally a fxed system desgn parameter 9

10 Adaptve Channel Reuse Tradeoff Large reuse dstance reduces CCI Small reuse dstance ncreases bandwdth allocaton Related work [Frodgh 92] Propagaton model wth path-loss only channel assgnment based on sub-cell compatblty [Horkawa 05] Adaptve guard nterval control specal case of adaptve channel reuse n TDMA systems Current work Propagaton models ncorporatng tme varaton of wreless channels statc (AWGN) channel, fast fadng and slow fadng Channel reuse n cooperatve cellular systems (network MIMO) compare wth sngle base staton processng 10

11 System Model Lnear cellular array, one-dmensonal, downlnk, sngle cell processng best models the system along a hghway [Wyner 1994] Full cooperaton leads to fundamental performance lmt More practcal scheme: adjacent 11 base staton cooperaton

12 Channel Assgnment Intra-cell FDMA, K users per cell, total bandwdth n the system K Bm Bandwdth allocated to each user maxum bandwdth Bm, correspondng to channel reuse n each cell may opt for a fracton of bandwdth, based on channel strength ncreased reuse dstance, reduced CCI & possbly hgher rate 12

13 Sngle Base Staton Transmsson: AWGN 13 Path loss only, receve power Pr ( d) A Pt d A: path loss at unt dstance γ : path-loss exponent Receve SINR d 0 APt 2 L 2L N d d L: cell radus. N0: nose power Optmal reuse factor Observatons arg max B m ( d, ) log 1 ( d, ) Moble close to base staton -> strong channel, small reuse dstance Reuse factor changes (1 -> ½) at transton dstance dt = 0.62 mle

14 Raylegh Fast Fadng Channel Envronment wth rch scatters Apples f channel coherence tme shorte than delay constrant Receve power P A g P g: exponentally dstrbuted r.v. r t d Lower bound: random sgnal Observatons Upper bound: random nterference AWGN and fast fadng yeld smlar performance reuse factor changes (1 -> ½) at transton dstance dt = 0.65 mle 14 Optmal reuse factor arg max B E log 1 ( d,, g) Both sandwched by same upper/lower bounds (small gap n between) m g

15 Raylegh Slow Fadng Channel Strngent delay constrant, entre codeword falls n one fadng state Optmal reuse factor arg max log 1 ( d,, g) B m Compare wth AWGN/slow fadng: optmal reuse factor only depends on dstance between moble and base staton Observatons Optmal reuse factor random at each dstance, also depends on fadng Larger reuse dstance (1/τ > 2) needed when mobles close to cell edge 15

16 16 Adjacent base staton cooperaton, effectvely 2 1 MISO system Channel gan vectors: sgnal nterference Transmtter beamformng optmal for solated MISO system wth per-base power constrant suboptmal when nterference present an ntal choce to gan nsght nto system desgn Base Staton Cooperaton: AWGN 2 2 ) 2 ( d L d h ,2 2 d L d L L I h ) ( ) ( ) ( j j j h h w w

17 Performance Comparson 17 Observatons no reuse channel n adjacent cell: to avod base staton servng user and nterferer at the same tme reuse factor ½ optmal at all d: suppressng CCI wthout overly shrnkng the bandwdth allocaton bandwdth reducton (1-> ½) overshadows beneft from cooperaton Advantage of cooperaton over sngle cell transmsson: only promnent when users share the channel; lmted wth ntra-cell TD/FD [Lang 06] Remedy: allow more base statons to cooperate n the extreme case of full cooperaton, channel reuse n every cell

18 Area Spectral Effcency BASE STATION A=.25D 2 p = S/I ncreases wth reuse dstance. For BER fxed, tradeoff between reuse dstance and lnk spectral effcency (bps/hz). Area Spectral Effcency: A e =SR /(.25D 2 p) bps/hz/km 2.

19 ASE wth Adaptve Modulaton Users adapt ther rates (and powers) relatve to S/I varaton. S/I dstrbuton for each user based on propagaton and nterference models. d S / d S Computed for extreme nterference condtons. Smulated for average nterference condtons. The maxmum rate R for each user n a cell s computed from ts S/I dstrbuton. For narrowband system use adaptve MQAM analyss

20 Propagaton Model Two-slope path loss model: S ( d) K d ( 1 d / g) S r a b Slow fadng model: log-normal shadowng Fast fadng model: Nakagam-m Models Raylegh and approxmates Rcean. ASE maxmzed wth reuse dstance of one! Adaptve modulaton compensate for nterference t,

21 Average Area Spectral Effcency [Bps/Hz/Km 2 ] ASE vs. Cell Radus 10 1 D=4R D=6R f c =2 GHz 10 0 D=8R Cell Radus R [Km]

22 Dstrbuted Antennas (DAS) n Cellular Basc Premse: Dstrbute BS antennas throughout cell Rather than just at the center Antennas connect to BS through wreless/wrelne lnks Performance benefts Capacty Coverage Power consumpton DAS

23 Average Ergodc Rate Assume full CSIT at BS of gans for all antenna ports Downlnk s a MIMO broadcast channel wth full CSIR Expected rate s C cst ( P) E u E sh log 2 1 S N I 1 f D( p, u) Average over user locaton and shadowng a 2 DAS optmzaton Where to place antennas Goal: maxmze ergodc rate p 7 p 6 p 2 p 1 p 3 p 5 p 4

24 Interference Effect Impact of ntercell nterference s the nterference coeffcent from cell j Autocorrelaton of neghborng cell codes for CDMA systems Set to 1 for LTE(OFDM) systems wth frequency reuse of one ), ( ), ( j N j j N u p D f u p D f SINR a a j

25 Interference Effect The optmal layout shrnks towards the center of the cell as the nterference coeffcent ncreases

26 Area Spectral Effcency Average user rate/unt bandwdth/unt area (bps/hz/km 2 ) Captures effect of cell sze on spectral effcency and nterference ASE typcally ncreases as cell sze decreases Optmal placement leads to much hgher gans as cell sze shrnks vs. random placement

27 Summary Wreless data/multmeda are man drvers for future generatons of cellular systems. Kller applcaton unknown; how wll cellular users access the Internet; wll cellular or WLANs preval. Effcent systems are nterference-lmted Interference reducton key to hgh system capacty Adaptve technques n cellular can mprove sgnfcantly performance and capacty MIMO a powerful technque, but mpact on outof-cell nterference and mplementaton unknown.

28 Dynamc Resource Allocaton Allocate resources as user and network condtons change Resources: Channels Bandwdth Power Rate Base statons Access BASE STATION Optmzaton crtera Mnmze blockng (voce only systems) Maxmze number of users (multple classes) Maxmze revenue : utlty functon Subject to some mnmum performance for each user

29 Dynamc Channel Allocaton Fxed channel assgnments are neffcent Channels n unpopulated cells underutlzed Handoff calls frequently dropped Channel Borrowng A cell may borrow free channels from neghborng cells Changes frequency reuse plan Channel Reservatons Each cell reserves some channels for handoff calls Increases blockng of new calls, but fewer dropped calls Dynamc Channel Allocaton Rearrange calls to pack n as many users as possble wthout volatng reuse constrants Very hgh complexty DCA s a 2G/4G problem

30 Varable Rate and Power Narrowband systems Vary rate and power (and codng) Optmal power control not obvous CDMA systems Vary rate and power (and codng) Multple methods to vary rate (VBR, MC, VC) Optmal power control not obvous Optmzaton crtera Maxmze throughput/capacty Meet dfferent user requrements (rate, SIR, delay, etc.) Maxmze revenue

31 Multcarrer CDMA Multcarrer CDMA combnes OFDM and CDMA Idea s to use DSSS to spread a narrowband sgnal and then send each chp over a dfferent subcarrer DSSS tme operatons converted to frequency doman Greatly reduces complexty of SS system FFT/IFFT replace synchronzaton and despreadng More spectrally effcent than CDMA due to the overlapped subcarrers n OFDM Multple users assgned dfferent spreadng codes Smlar nterference propertes as n CDMA

32 Rate and Power Control n CDMA * Optmze power and rate adaptaton n a CDMA system Goal s to mnmze transmt power Each user has a requred QoS Requred effectve data rate *Smultaneous Rate and Power Control n Multrate Multmeda CDMA Systems, S. Kandukur and S. Boyd

33 System Model: General Sngle cell CDMA Uplnk multple access channel Dfferent channel gans System supports multple rates

34 System Model: Parameters Parameters N = number of mobles P = power transmtted by moble R = raw data rate of moble W = spread bandwdth QoS requrement of moble, effectve data rate, s the R ( 1 P e )

35 System Model: Interference Interference between users represented by cross correlatons between codes, C j Gan of path between moble and base staton, L Total nterferng effect of moble j on moble, G j s G L C j j

36 SIR Model (neglect nose) SIR j G G j P P j E I b o SIRW R

37 QoS Formula Probablty of error s a functon of I Formula depends on the modulaton scheme Smplfed P e expresson QoS formula e c P 1 e R SIRW P R 1

38 Soluton Objectve: Mnmze sum of moble powers subject to QoS requrements of all mobles Technque: Geometrc programmng A non-convex optmzaton problem s cast as a convex optmzaton problem Convex optmzaton Objectve and constrants are all convex Can obtan a global optmum or a proof that the set of specfcatons s nfeasble Effcent mplementaton

39 Problem Formulaton Mnmze 1 T P (sum of powers) Subject to R 1 R P 0 P e R thresh SIRW R Can also add constrants such as P P P P mn max

40 Results Sum of powers transmtted vs nterference

41 Results QoS vs. nterference