Lecture 6 Admission control. Admission control

Transcription

1 Lecture 6 The task of the admission control is to Predict the impact of adding new user(s) to the quality of service of the currently active connections Predict the resource consumption of the new user(s) Based on the above, decide whether the new user(s) can be admitted to the system so that They achieve requested QoS The QoS of active calls stays above minimum required level S Radio Resource Management Methods 3 op TKK Comnet 2

2 errors Type I New mobile is erroneously accepted and the system becomes congested (QoS of an active call drops below minimum tolerable level) Type II New mobile is erroneously reected. The call was blocked even though there wod have been enough resources in the system to accommodate the user. Type I errors lead to high outage and low QoS, type II errors lead to high blocking and low resource utilization (low capacity). S Radio Resource Management Methods 3 op TKK Comnet 3 Simple admission control scheme. The number of channels in the system is fixed The system monitors the number of occupied channels New call is admitted if there is free channels; otherwise the call is blocked. Stochastic admission control Instead of assigning new channels automatically to incoming calls, the call is admitted with probability p i when there is i calls in progress. The probability is utilized to priorize handoff calls over new calls. S Radio Resource Management Methods 3 op TKK Comnet 4 2

3 Power based admission control In power controlled systems, adding one user to the system will affect the transmit power levels of all the others. New user is admitted if the associated power control problem is feasible. That is, all the users can achieve their target SINRs. Otherwise, the call is reected. Centralized admission control scheme. Let H denote the normalized link gain matrix corresponding to the active calls. The system is assumed to be feasible, hence ρ(h)<. Let H denote the normalized link gain after new user is admitted. The matrix H is a principal submatrix of H. Thus ρ(h )>ρ(h). New user is admitted only if ρ(h )<. In practice the feasibility of the power control problem cannot be checked before the user is admitted and the evolution of power is observed, since it wod be very diffict to estimate H. S Radio Resource Management Methods 3 op TKK Comnet 5 Threshold based admission control Simple threshold based admission control schemes estimate the network load by measuring the total received power. If the power is below some threshold value, then new user is admitted; otherwise it is reected. The performance of threshold based admission control schemes is very much dependent on how the threshold is set. S Radio Resource Management Methods 3 op TKK Comnet 6 3

4 Threshold based admission control (Andersin and Rosberg, 997) S Radio Resource Management Methods 3 op TKK Comnet 7 Probing / Iterative admission control New user is gradually admitted to the user by ramping up its power. To limit the impact of the new user on the active ones, their power levels can be initially increased. Power control is conducted and the QoS of the users observed. If the QoS of the active users stay large enough during the admission period and the new user achieves its target QoS, the user is admitted. If the QoS of some of the active users decrease below some protection level, new user is reected. S Radio Resource Management Methods 3 op TKK Comnet 8 4

5 Soft and safe admission control (Andersin 997) Assume that initially user are i=,2,,m are transmitting in the system. Assume further that the associated power control problem is feasible and the initial power values correspond to the solution of the power control problem N * * i i ηi i = ì * p = hp() + p, =, 2,..., M ( ) p = I H η p Assume now that user is trying to get access to the system. S Radio Resource Management Methods 3 op TKK Comnet 9 Soft and safe admission control (Andersin 997) Now the associate power control problem becomes N p = hp+ h p+ η, =,2,..., M i i i i = ì N p = h p + η i i = Mobile can be safely admitted if there exist solution to the above power control problem such that pi pi for all users. S Radio Resource Management Methods 3 op TKK Comnet 5

6 Soft and safe admission control (Andersin 997) Assume that the power of the new mobile is bound to ( ) ( ) p μ ξ N η i ξ = min i M h i ( μ() )( ) η ( μ( ) ) ( () ) p () = h p + h p () + η h p () + μ + η i i i i i i = = ì ì p () + I H = + p * * N The maximum power value for the new user that the existing users can cope with is p i μ () min i N * pi () S Radio Resource Management Methods 3 op TKK Comnet Soft and safe admission control (Andersin 997) Iterative admission control. Initial conditions p * ()=(,p *,p 2*,,p M* ) T η i ξ = min i M h i 2. Set k:=k+ 3. If p () = then reect mobile and stop. 4. Set p ( k ) p 5. If μ(k)=, reect mobile and stop. Otherwise, set p( k) p( k ) ( + μ( k) ) + μ( k) ξ p( k) = p( k) * pi( k) = pi ( k ) ( + μ( k) ), i=,2,... M 6. Run power control * p( n+, k) = H' p( n, k) + η ' p ( k) 7. If Γ(p * (k))>γ t and accept the mobile p( k) p Otherwise goto 2 p i = min pi ( k ) μ ( k ) i N * S Radio Resource Management Methods 3 op TKK Comnet 2 6

7 Soft and safe admission control (Andersin 997) S Radio Resource Management Methods 3 op TKK Comnet 3 Active link protection (Bambos 2) Active link protection (ALP) access control scheme for uplink New user is allowed to increase its power gradually. If SIR is not increasing fast enough, the user is reected. By properly chosen the step size active link protection property can be assured. I.e. new user can not cause outage to old already accepted user. The interference caused by new users is compensated by using higher SIR-target for the active users (Protection ratio). Several users can be handled at the time. The ALP call admission control scheme (ALP-CAC) is free of type I errors but suffers from Type II errors due to fixed protection ratio. S Radio Resource Management Methods 3 op TKK Comnet 4 7

8 Active link protection (Bambos 2) Power control: W gki pi ( n) Γ i( n) =, i B R g p ( n) + N W i k = pi ( n), i S pi ( n+ ) = δ δ pi ( n), i O δ > Power control step size t { i i} t { i i} Sn ( ) = i: Γ ( n) mγ On ( ) = i: Γ ( n) < mγ m > SIR-margin k Set of supported users Set of users in outage SIR of user i bit bang-bang power control S Radio Resource Management Methods 3 op TKK Comnet 5 Active link protection (Bambos 2, Jäntti 2) If the number of users simtaneously trying to access 2 the system is limited to M new, SIR-margin is set to m δ and the initial power of new users is upper bounded by δ p < NW M g new max g max where is the maximum link gain, then t i S( n) Γ i( n+ ) Γi i O( n) Γ ( n+ ) >Γ ( n) i i That is, if SIR of a user is above the SIR-target at a given time instance, then SIR will stay above the target in the future regardless of the users trying to get access. Furthermore, new users are guaranteed to have increasing SIR. S Radio Resource Management Methods 3 op TKK Comnet 6 8

9 Active link protection (Bambos 2) S Radio Resource Management Methods 3 op TKK Comnet 7 in CDMA systems 9

10 Uplink SIR W Pi Γ i = υiri Itot Pi Received power Pi = I W + υ RΓ i i i tot Interference + noise power Load factor I = ( + ι) P+ NW= ( + ι) LI + NW tot tot Iext ι =.55 Other cell to own interference ratio I own W = 3.84Mcps Chip rate Ri υi.67 υ = i Li = W + υ R Γ i i i Bit rate Voice activity Activity factor for data calls Uplink load factor Noise rise Itot Itot = = NW I ( + ι) LI ( + ι) L tot tot S Radio Resource Management Methods 3 op TKK Comnet 9 Uplink load Load factor Itot Itot = = = NW I ( + ι) LI ( + ι) L η η = ( + ι) tot tot L η Pole capacity Pi = Li NW, η η Maximum load, if power η = ( + ι) L = constraint is relaxed S Radio Resource Management Methods 3 op TKK Comnet 2

11 Load factor Let us consider only single service class η = ( + ι) L = ( + ι) NL Maximum number of users for certain service ( + ι) η = N W + υrγ W η N = + υrγ ( + ι) N Number of simtaneous users Maximum rate for given number of users η W ( + ι) N W If >>, then R = υrγ η υγ ( + ι ) N R ( + ι) Nυ Γ η W Wη NR ( + ιυ ) Γ S Radio Resource Management Methods 3 op TKK Comnet 2 Transmission power Li Li pi = NW NW g η g η i i,max Noise rise Noise rise Noise rise tells how much larger power is needed due to mtiuser communication compared to single link Power constraint pi pmax, i=,2,..., N vs η η, max It is easier to set limit to the relative power change than to consider the individual power constraints S Radio Resource Management Methods 3 op TKK Comnet 22

12 Noise rise Noise rise as a function of load I tot NW Noise rise (db) Load η Typical operation range 3dB db Typical operation range.5.9 S Radio Resource Management Methods 3 op TKK Comnet 23 Capacity vs coverage Noise rise limits the capacity. Pole capacity can not be exceed even if there wod be infinite amount of power available By increasing the load η the coverage area is decreased and vice versa Cell breathing: Cell size is allowed to vary according the load. P d η η Li gi = NW p η i S Radio Resource Management Methods 3 op TKK Comnet 24 2

13 Uplink Statistical-based admission control New user is granted access to the network if the load with it stays below some prefixed load limit η. It is assumed that only the load in own cell is changed. Hence, the other-to-own cell interference ratio ι is assumed to change. This assumption implies that there is no need to estimate ι. Mtiply by +ι if other cell interference is also assumed to change due to the admission of new user η +Δ L < η, th, Δ L = W + υrγ υr ηdl +Δ Ldl < ηdl, th, Δ Ldl = ( α) Γ W Uplink Downlink S Radio Resource Management Methods 3 op TKK Comnet 25 Wideband Power-based Uplink New user is granted access to the network if the total received power stays below prefixed limit Itot +Δ Itot < Itot,max I tot,max Total received power can be easily measured ΔI tot must be estimated or measured: New user is allowed to gradually increase its power. If I tot increases above the set maximum value, new user is dropped (ALP Active Link Protection) If load limit is given, I tot,max can be found Noise rise I = I = N W tot S Radio Resource Management Methods 3 op tot,max TKK Comnet NW 26 η η, th 3

14 can be estimated from the derivate ΔI tot ditot d NW NW = = dη dη η η ( ) ΔI di N W Δη η η tot tot =, Δ η 2 =Δ d ( ) NW I Δ Δ = Δ tot Itot L L 2 ( η ) η 2 L I tot I tot,max ΔI tot ΔL η Optimistic estimate: Lower bound for ΔItot S Radio Resource Management Methods 3 op TKK Comnet 27 can be estimated from the integral ΔI tot η +Δ L η +ΔL NW tot tot η 2 η ( ) η η Δ I = di = d NW NW ΔL NW ΔL I η ΔL η η ΔL η η ΔL = = = Alternatively NW NW Δ Itot = η ΔL η ΔL NW ΔL I η ΔL η η ΔL = = tot tot S Radio Resource Management Methods 3 op TKK Comnet 28 4

15 In practice, a convex combination of the derivative and integrate methods is utilized Itot L I α tot L ( ) Itot, α Δ Δ Δ + η η L α Δ Δ L = W + υrγ S Radio Resource Management Methods 3 op TKK Comnet 29 The admission threshold I tot,max can be predefined based on the planned noise rise, or it can be dynamically adusted based on the realized noise-rise and other-to-own cell interference ratios. I ( t) = f( NR( t ), NR( t 2),...) tot,max By decreasing I tot,max the interference caused to other cells decreases. Also the smaller I tot,max the smaller the Noise rise. For instance, an I-controller cod be utilized to control the threshold based on the realized noise rise. S Radio Resource Management Methods 3 op TKK Comnet 3 5

16 Threshold for maximum uplink load is controlled based on outage probability. Other-to-own cell interference ratio is random Target outage.5 η Iteration Outage probability Iteration S Radio Resource Management Methods 3 op TKK Comnet 3 and handovers Interference Guard Margin (IGM) is an extension of the guard channel principle to CDMA. New user is admitted to the cell if Itot + Δ Itot < Itot,max IGM I tot Handoff call is admitted to the cell if Itot + Δ Itot < Itot,max New calls blocked only handoff calls are admitted I tot,max Itot,max IGM New calls can be admitted S Radio Resource Management Methods 3 op TKK Comnet 32 6

17 Downlink Based on available power In downlink, transmitted power is used instead of received power. Ptot +Δ P< Pmax Based on available code resources S Radio Resource Management Methods 3 op TKK Comnet 33 IS-CDMA systems, the resources to be allocated among the users are base station power and channelization codewords. In order to minimize the intra-cell interference, orthogonal codes can be utilized. An example of an orthogonal code is the Walsh code which can be generated by using a simple matrix recursion {} {,} {,-} {,,,} {,,-,-} {,-,,-} {,-,-,} C C C, C k k k = = Ck C k S Radio Resource Management Methods 3 op TKK Comnet 34 7

18 For instance, if the spreading factor n = 4 the number of orthogonal codes is also 4. Users may require different data rates and thus need different spreading factors. In this case the allocated codes need to satisfy a prefix condition. The prefix condition requires that for a codeword c n of length n and elements (c, c2, cn) there is no other code of length m < n having elements (c, c2,, cm) for m n. Orthogonal variable spreading factor (OVSF) code is a code that allows mtiplexing users with variable spreading factors into same channel while keeping the codewords uniquely decodable and orthogonal. S Radio Resource Management Methods 3 op TKK Comnet 35 Let N denote the number of users sharing the channel. We wod like to assign a uniquely decodable codewords of length n i for each user i =, 2, N. The Kraft s inequality states that the necessary and sufficient condition for the existence of such code is where a denotes the size of the alphabet. For binary codes a = 2. The spreading factor is given by S i =a n i=w/r i Thus The Kraft s inequality becomes n n ri W SF = = S Radio Resource Management Methods 3 op TKK Comnet 36 8

19 When checking the availability of codes, one shod take into account that part of the code tree cod be reserved for common and shared channels {,,,} {,} SF SF D CS D set of dedicated channels CS Common and shared channels {} {,-} {,,-,-} {,-,,-} {,-,-,} S Radio Resource Management Methods 3 op TKK Comnet 37 9