Hosein.tt

5012 Wateridge Vista Dr., San Diego, CA 92121, U.S Email: {patrick.hosein, rath.vannithamby}@ericsson.com solved by providing all relevant data to a centralized func-tion that computes the allocation of BSs to Mobile Sta- tions (MSs). However such an approach would be verysignaling intensive. We instead focus on a distributed so- Recent developments in wireless networks have focused on lution in which each mobile periodically determines if it improved data performance (e.g., 1xEV-DV [5], 1xEV-DO should switch from its presently assigned BS (its serving [1]and HSDPA). In these networks the forward packet data BS) to any of the other BSs in its active set. This decision channel is time shared among the data users. In order to is based on local information as well as any information be able to take advantage of fast fading effects (i.e., sched- provided to it by the BSs in its active set.
ule users when they are in good radio conditions), the This approach actually improves the overall system fair- scheduling of users is performed at the Base Station (BS).
ness because of the following. Consider any MS and as- This implies that it is not possible to support soft handoff sume that it is presently being served by BSs and is con- on this forward packet data channel since this would re- sidering switching to BSt. Assume that we can determine quire that the BSs synchronize their schedules. However, the forward link throughput of the mobile if it were to macro-diversity gains can be achieved by fast cell switch- switch cells. Let rs denote the present throughput of the ing. The mobile is served from the base station with the MS and let rt denote the throughput that the MS will best forward link quality. This BS is reported through the achieve if it were to switch to cell BSt. If rs < rt then the reverse channel quality indicator channel. However, the mobile can increase its forward link throughput by switch- BS with the best forward link quality is not necessarily ing. Furthermore, since rs < rt, then the loading of BSs the best BS to serve the mobile since it may be congested.
is higher than that of BSt. If the MS makes the switch In this paper we describe a procedure whereby a mobile then the loading on BSs decreases while the loading on switches to the cell that can provide it with the high- BSt increases. Hence the switch reduces the difference est throughput rather than best channel conditions. We in the loading between the two cells. Therefore, the sys- demonstrate why this approach is preferable and describe tem settles to a state where all cells running near capacity the framework necessary to support this mechanism.
experience comparable loading. Note that the same argu-ment can be used for the reverse link loading. Both for- ward link and reverse link loading must be considered indeciding if a MS should switch since link asymmetry may In a distributed system such as a wireless network, the re- result in one increasing in throughput while the other may sources available to each network component (e.g., BS) must be efficiently allocated among the correspondingusers of the component. In addition, the loading of each Cell switching should not be performed too often because component must be optimally balanced so as to maximize of the signaling overhead it induces.
network resource utilization. We consider the problem of throughput of a MS depends on the amount of power al- optimal balancing of load (data users) among BSs. Note located to the F-PDCH, the number of Walsh codes al- that each mobile induces BS loading in both the Forward located to the F-PDCH, the number of active forward Packet Data Channel (F-PDCH) and Reverse Packet Data link users in the cell, the forward link radio conditions Channel (R-PDCH). In this paper we address the problem of these users, and the scheduler (which defines the trade- of forward link loading but a similar approach can be used off between fairness and sector throughput). Each of these for the reverse link. Note that loading in both directions factors need to be considered when predicting the achiev- must be computed before making a decision to switch.
able rate of the MS at the target cell. We first describethe present procedure for cell switching and highlight the The optimal allocation of users among the BSs can be problems with the approach. We next provide a method for predicting the throughput experienced by a MS if it MS is assumed to report the channel quality from the sec- were to switch to another BS. We demonstrate that the tor of the strongest pilot with the sector’s identification.
BS that provides the best throughput need not be the onewith the best forward link channel conditions. We then discuss some practical and standardization issues.
In any cellular wireless system, the MS may move out ofthe coverage of any particular serving BS or the channel condition changes due to environment changes. A processby which the MS indicates to the BS that a different sector This section describes the current procedures for cell should now be its forward link serving sector is, therefore, switching in the 1xEV-DV standards. We first describe necessary. The main goal of the cell switching process is the channel that is used to convey the information indi- to allow the best sector to serve the MS and make the cating that a cell switch is needed. Next we describe the transition robust. The MS sends a distinct switching pat- procedure to determine if a MS should switch sectors and tern in the R-CQICH for a period of time known as the how this request is indicated to both the source and tar- cell switching period. The length of the switching period get sectors. We then discuss some of the limitations of the is configurable and is dependent upon whether the source present standards. In particular the fact that more infor- and target sectors belong to a single BS. The BS, further- mation needs to be taken into account for such a decision.
more, has the capability to terminate the switching period In the final section of this paper we suggest possible ways before the expiration of the configured switching period by for the dissemination of this information.
sending a message to the MS on the F-PDCCH. At thetermination of the switching period, the MS resumes non- cell switching (normal) operation of the R-CQICH with The Reverse-Channel Quality Indicator Channel (R- the selected target sector. During the cell switching pe- CQICH) is a new reverse link channel added to support riod, the slots at the end of the frame are punctured and the fast link adaptation of the forward packet data channel replaced with switching slots containing fixed data covered (F-PDCH) in 1xEV-DV systems. Link adaptation allows by the Walsh cover of the target sector whereas preced- the BS to track changes in the channel quality and se- ing slots carry the Walsh cover of the source sector. The lect the optimal data rate. The MS subscribing to the F- switching slots are punctured at the end of the frame to PDCH service has to transmit an indication of the forward minimize their impact upon the accuracy of CQI reports link quality in the R-CQICH. This indication of quality is and the data is fixed to improve the detection of the target the measured pilot strength seen at the MS from the best BS in the MS’s active set that contains all the pilots that The number of switching frames and number of switching the MS is in soft or softer handoff. Since the channel qual- slots can be configured. When the number of switching ity is only that of the strongest pilot seen at the MS, the frames is chosen to be large then the number of switching MS must also indicate the identity of that best BS in the slots needed are small, and vice versa. Other factors such R-CQICH. Based upon the information on the R-CQICH, as repetition to increase the reliability of the cell switch the BS determines the data rate of the F-PDCH and sends detection and control hold generate the need for different the data rate to the mobile via an explicit overhead mes- sage on the forward link using the F-PDCCH.
Each frame (20 ms) is divided into 16 slots of 1.25 ms du-ration. The first slot contains the full CQI report and the In order for the MS to perform a meaningful cell switching following slots contain the differential CQI reports repre- process, basically, it needs to know whether it can receive sented by 1 bit or full CQI reports represented by 4 bits in better performance (throughput, delay, etc.) if switched differential or full CQI reporting mode respectively dur- to another sector in the active set. In order to know this, ing normal operation. The reporting pattern can change the MS needs information such as pilot strength, sector during cell switching, control hold and link imbalance.
load, resource availability such as available Walsh codesand transmit power, and the type of scheduling algorithm used at the scheduler in the current and target sectors.
Soft handoff is not supported for F-PDCH, instead macro Unfortunately, the MS knows only the pilot strength of diversity in the F-PDCH is achieved through cell selection.
current and the target sector; the rest are unknown at The F-PDCH is not soft-combined at the MS, the MS in- the mobile station. Currently, in the 1xEV-DV systems, dicates to the BS, on the R-CQICH, the sector from which the MS performs cell switching purely based on the pi- the mobile station wishes to receive the F-PDCH. The R- lot strength. The 1xEV-DV standard does not limit the CQICH indicates the desired sector by selecting a length-8 MS in its cell switching decision except that the BS can Walsh code (or R-CQICH cover) for Walsh spreading of provide the MS with an estimate of the delay and hence coded symbols. Consequently, the cell selection informa- the degree of the service interruption if the MS were to tion for a particular F-PDCH is also transmitted on the R-CQICH besides the channel-quality information. The The standards also allow a mechanism to reject cell switch- ing. It is performed by sending a special control message on the F-PDCCH. However, without knowing the needed information previously mentioned, the MS cannot performan intelligent cell switch. As a result, the user may faceunnecessary degradation in performance and the network may experience imbalance in load consequently leading tocongestion.
In this section we propose a procedure for each MS to de-termine its best serving sector defined as the sector that provides the highest throughput in the forward link. On a periodic basis each BS broadcasts two parameters thatrepresent the BS’s loading information. Each BS uses this Figure 1: Scheduler attenuation factor versus number of information together with other local information to esti- mate the throughput it can achieve if it were to switch toany of the other BSs in its active set. If it can increaseits throughput significantly by switching then it executes a switching procedure. First we compute the predicted forward link throughput for an arbitrary BS/MS pair.
Note that in the case of a Round Robin scheduler this at-tenuation factor is simply N/(N + 1). However, for Pro- Assume that a Proportional Fair scheduler is being used portional Fair scheduling the user diversity gain increases for the F-PDCH (a similar approach can be used for other with the number of users and hence the attenuation func- schedulers). One can show that for each user i, the ratio tion decreases at a slower rate (see Figure 1 for attenu- of the user’s average forward link throughput r ation plots for both Round Robin and Proportional Fair schedulers). Note that this attenuation factor is strictly [2]). Since this constant is the same for all MSs, we can dependent on N and can therefore be computed for var- update our estimate of it each time a new frame (not a ious values of N and stored in a table. The attenuation retransmission) is transmitted. Let us index these new factor γ(N ) effectively takes into account the dependence transmissions with k then our estimate of θ is obtained as on the number of active users in the system.
We now use this parameters to compute the expected value of the rate at which the MS would be served at thetarget cell. The served rate of the MS is determined by its where r(k) is the average throughput and d(k) is the av- channel quality (denoted by q, the SINR of the pilot chan- erage served rate of the user who is scheduled in the time nel), the transmission power of the F-PDCH (denoted by slot corresponding to index k. Note that we are in fact P ) and the number of Walsh codes used for the F-PDCH averaging this quotient over all users being served since denoted by W . The BS will have some look-up table that they all converge to the same value. This increases the maps these three quantities to a rate. Since Turbo codes rate of convergence. This ratio will allow us to estimate are used we can estimate the served rate d by the Shannon the throughput given the achievable rate of the MS.
This provides the present estimate of θ but the addition of a new user to the cell will cause this metric to change.
We next estimate the effect of a new user on this metric.
One can show (see [3] for details) that for the case of Pro- where p is the transmission power of the pilot used by portional Fair scheduling, the throughput of an individual the MS to determine q and B is the channel bandwidth.
user varies with the number of users in the system by the We can combine this with the estimate of θ as well as the attenuation factor due to an increase in the number ofusers to obtain the predicted rate as Therefore, if the MS were to switch, its throughput wouldbe decreased due to the increase (by one) of the number Both α and β are BS specific but independent of the MS.
of users in the target cell by the following multiplicative Note that all relevant information is accounted for, the scheduler, the number of active users, the number of avail- able Walsh codes, the power available to the F-PDCH andthe only MS specific dependence, the pilot SINR, q. We assume that each BS periodically broadcasts its most re-cent estimate of α and β. Each MS can then use the aboveexpression to determine its predicted throughput if it were to move to another cell. We assume that BS j broadcastsαj and βj every τ seconds (probably on the same broad- cast channel that is used to broadcast the reverse link loading). We also assume that every T seconds each MSdetermines the forward link channel quality qj for each BS j. Let A denote the active set of the mobile. Every T seconds each MS determines the predicted throughput foreach BS in its active set (including its serving sector) and then determines the cell j∗ that can potentially provide Figure 2: Predicted Throughput as a Function of Pilot Note that MSs will typically decide to switch near the celledge. If this is the case then q is small and we can make the approximation log (1 + x) ≈ x to obtain In this case only a single parameter needs to be broadcastby the BS. Note that this approach will tend to give the channel quality of a MS higher weighting over the otherfactors and hence may be preferable in some situations.
We assume that the BS broadcasts are much more fre-quent than the MS cell switching determination and that each MS performs this operation at time instances thatare uncorrelated with those of other MSs. In this way the effects due to the simultaneous switching of two MSs into a cell are minimized. This effect is also minimized since wewill only allow a MS to switch if its predicted throughput Figure 3: Predicted Throughput as a Function of Pilot in the new cell is significantly larger (not simply larger) SINR for different Numbers of Walsh Codes In this section we provide some illustrative examples to still be better, from a throughput perspective, to have the demonstrate that the proposed approach in fact results in MS switch to the 5 user BS because of its lower loading.
a more efficient network as opposed to the present proce- Next we consider what happens if there are differences in dure based solely on channel conditions.
the number of Walsh codes available to the F-PDCH. We The ratio θ will vary with the number and distribution of consider the case of 10 users and keep everything identical users in the sector. Typically it will decrease monotoni- except that one BS has 16 available Walsh codes for the cally with the number of users in the system. The next F-PDCH while the other has 32. In this case (Figure 3) factor is the attenuation factor γ(N ). Consider two BSs the 32 Walsh code BS provides better throughput but the that only differ in the number of users. In one case there are 5 users and in the other there are 10 users. We deter- Finally we consider the effect of the power available to mine the expected value of the product θγ(N ) for these the F-PDCH. We repeat the case of 10 users but this time two cases and plot (in Figure 2) the resulting predicted we assume that one BS has 14 watts available to the F- throughput as a function of the channel conditions (SINR PDCH while the other has 7 Watts. Here we see (Figure 4) a significant throughput difference between the two cases Note that even if the forward link channel conditions for and so it is preferable to have the MS switch to the BS the 5 user BS is worse than that of the 10 user BS it may that the overhead will be high because of the mobile spe- The sector information could be distributed among other sectors via the backhaul connections, and each sector putsa subset of the information of itself and its neighboring sectors in a F-PDCCH broadcast message. If a single F- PDCCH message is used, the amount of information that can be conveyed may be too little, on the other hand, if multiple F-PDCCH messages are used, then there could be a long interruption in serving MSs.
The sector information could be distributed among other sectors via the backhaul connections, and each sector broadcast this information to all the MSs in the sector via the F-PDCH with a low rate to maintain high relia-bility. This in-traffic broadcast via the F-PDCH channel Figure 4: Predicted Throughput as a Function of Pilot can be made possible through a common identifier in the associated F-PDCCH message. This allows for sufficientgranularity in representing the sector information of eachneighbor sectors.
In this section we describe a simple algorithm for cell switching using the above procedure for determining theexpected throughput when switched. One simple embod- We considered the problem of assigning MSs to their serv- iment of the algorithm is as follows. On a periodic ba- ing BSs. We described the present distributed procedure sis the MS determines j∗. If this is the serving BS then for making these assignments and show that, in terms of nothing is done. If this is a non-serving BS and the pre- system data throughput, the procedure was non-optimal.
dicted throughput is at least ε greater than the present We then presented an approach that switches MSs based throughput and the predicted reverse link throughput is on the throughput that the MS is expected to experi- also greater than the present reverse link throughput then ence if it were to switch. In this way each MS is always served by the BS that can provide it with the highestexpected throughput. We provided illustrative examples Note that the estimate of the forward link throughput of that demonstrate why the proposed approach was better the serving BS is used in the algorithm and not a smoothed than the present procedure. Finally we described various measurement of the actual throughput. The latter quan- options for disseminating the information that is required tity can in fact be used to fine tune the algorithm to obtain for the operation of the proposed procedure.
[1] P. Bender et al., “CDMA/HDR: A bandwidth- Recall that two parameters (α and β) must be reported to efficient high-speed wireless data service for nomadic the MSs in a periodic fashion. We provide several (non- users”. IEEE Commun. Mag., pp 70–77, July 2000.
standardized) alternatives for accomplishing this in thefollowing paragraphs.
[2] P. Hosein, “Capacity Model for the CDMA/HDR High Speed Wireless Data Service”, in Proc. of the The sectors could periodically send the sector information 4th ACM International Workshop on Modeling, Anal- via the F-PDCCH message. This is a broadcast message ysis and Simulation of Wireless and Mobile Systems, that all MSs can monitor. The MS knows the predefined time instance that the serving sector and the other sec-tors in the active set will send this information. Hence, [3] P. Hosein, “A TCP-Friendly Congestion Control Al- the MS can acquire this information from its serving and gorithm for 1XEV-DV Forward Packet Data”, in other sectors by locking onto their transmission at the Proc. Spring VTC 2003, Jeju, Korea, 2003.
predefined times slots. The disadvantage of this method [4] S. Lu, et. al., “Fair scheduling in wireless packet net- is that multiple RF channels need to be decoded at the works,” in Proc. IEEE INFOCOM,, 1998, pp. 1108- The sector information could be distributed among other [5] A. Soong et al., ”Forward High-Speed Wireless sectors via the backhaul connections, and each sector Packet Data Service in IS-2000 (1xEV-DV)”, IEEE places the necessary information in a MS directed F-

Source: http://hosein.tt/papers/iee2004-rath.pdf