On Self-Optimization of the Random Access Procedure in 3G Long Term Evolution

Mehdi Amirijoo, Pl Frenger, Fredrik Gunnarsson, Johan Moe, Kristina Zetterberg

Wireless Access Networks, Ericsson Research, Ericsson AB, Sweden.
{mehdi.amirijoo, pal.frenger, fredrik.gunnarsson, johan.moe, kristina.zetterberg}@ericsson.com

AbstractOperationally efficient radio networks typically feature a high degree of self-organization. This means less planning efforts and manual intervention, and a potential for better radio resource utilization when network elements adapts its operation to the observed local conditions. The focus in this paper is selfoptimization of the random access channel (RACH) in the 3G Long Term Evolution (LTE). A comprehensive tutorial about the RACH procedure is provided to span the complexity of the selfoptimization. Moreover, the paper addresses RACH key performance metrics and appropriate modeling of the various steps and components of the procedure. Finally, some coupling between parameters and key performance metrics as well as selfoptimization examples are presented together with a feasibility discussion. The main ambition with this workshop paper is to present and define a relevant set of self-optimization problems, rather than to provide a complete solution. Keywords 3GPP; LTE; Self-Organization; Random Access;RACH; E-UTRAN; Self-Tuning; Self-Optimization

Self-healing, i.e. algorithms to handle disruptive events and to minimize negative consequences on services.

I.

INTRODUCTION

There is a strong momentum for Self-Organizing Network (SON) features in wireless communication networks, both via requirements from operators and through standardization work. The Next Generation Mobile Network (NGMN) association of operators brings forward requirements on management simplicity and cost efficiency [1]. The vision is that algorithms automate tasks that currently require significant planning efforts. In parallel, the 3rd Generation Partnership Program (3GPP) works on specifications for 3G Long Term Evolution (LTE), and SON is central in the network management and optimization discussions [2]. SON can mean vastly different things, but three components are central Self-configuration, i.e. plug and play functionality where network elements are configured (identity allocation, software upgrade, communication link establishment, etc) automatically. Self-optimization, i.e. a more or less continuous adaptation of parameters to meet specified requirements, typically specified at a high level.

The random access procedure in LTE can benefit from selfoptimization. A mobile (User Equipment, UE in LTE) in an idle state is essentially unknown to the network (except for some rough localization information). In order to establish a relation to the network, for example to initiate some service, the mobile scans the carrier frequencies to determine the most suitable cell and associated base station (eNodeB or eNB in LTE) for communication. The broadcast information from this base station provides the mobile with cell-specific random access procedure details. Optimal random access performance is central to obtain intended coverage and low delays, while avoiding excessive interference to communication links in other cells and maintaining a desired balance in the radio resource allocation between random access and data services.. The considered delays include call setup delays, session resuming delays, handover delays, etc. The challenge is to balance the resource allocation between random access and other communication needs, while adapting to local radio characteristics, cell size and variations in terms of traffic in the cell and neighboring cells. One approach to random access procedure configuration is to use a set of standard parameter values in all base stations, typically based on extensive simulations. This may, however, result in a suboptimal performance since the cell-specific characteristics are not catered for. Another approach is to by means of simulation, prediction, or field trials evaluate a wide range of random access parameters and choose those cell-specific settings that satisfy given requirements. The drawbacks include the need for extensive simulation, planning and/or field trial efforts. Furthermore, it is difficult to be responsive to variations in the radio network, not the least due to gradual deployment of additional network elements. Therefore, self-optimization of the random access procedure has great potential. Self-organization and tuning have been previously addressed in the literature. For an overview on autonomic communication in networks refer to [3]. Automation of neighbor relation lists has received some attention lately [4][5]. Several publications related to automation in 3G networks exists, e.g., capacity and coverage balancing [6][7], and admission control [8]. The project

978-1-4244-3924-9/09/$25.00 c 2009 IEEE

29

SOCRATES aims at the development of self-organization methods for future wireless access networks [9]. Papers [10][11] address WCDMA random access optimization by means of simulations. The outline of this paper is as follows. Section II gives an extensive tutorial on the random access procedure in LTE, while Section III addresses relevant performance specification aspects and modeling. Selected experiments in Section IV illustrate the relation between random access parameters and traffic on one hand and performance of the other, and Section V gives a selfoptimization example, before Section VI concludes the paper. II. RANDOM ACCESS PROCEDURE IN LTE
TX

TX

CP
103 s

RA sequence
839 samples = 800 s

time

a) Format 0

CP
684 s TX

RA sequence
800 s

time

b) Format 1

CP
203 s TX TX

RA sequence
1600 s

RA sequence

time

c) Format 2

The random access procedure in LTE is performed at any of the following five events: i) initial access of an idle mobile, ii) reestablishment after radio link failure, iii) handover to a different cell, iv) downlink data transmission to a mobile, which is out of time-synchronization, and v) uplink data transmission from an out-of-synch mobile. In all cases, one objective is to establish uplink time synchronization, while in some it also provides the means for the mobile to notify the network about its presence, and for network to give the mobile initial access. At events iii) to and v), the serving base station can control the procedure to avoid collisions and ambiguities in the random access (non-contention based procedure). However, in the general case, the possibility of a collision, or contention, between different users access attempts needs to be handled (contention-based procedure). The former is essentially a simpler version of the latter. Prior to sending the random access preamble, the mobile performs cell selection if necessary, and establishes downlink synchronization. The mobile acquires broadcasted information about the random access resources and procedure configuration. These parameters are further described in the following subsections. For further details on the random access procedure in LTE, see [12][13][14][15][16]. A. Random Access Physical Resources The random access physical resource consists of a set of preambles, a set of formats, and a set of random access opportunities. 1) Random Access Preambles The requirements on the sequence comprising the preamble are two-fold: good correlation properties to allow precise arrival time estimation and low correlation with other preambles to suppress interference from other mobiles. A sequence that has ideal such properties is the Zadoff-Chu sequence (root sequence) [14][17]. The periodic auto-correlation function (ACF) of a Zadoff-Chu sequence is only non-zero at time-lag zero (and periodic extensions) and the magnitude of the correlation with other sequences is equal to the square-root of the sequence length N. In LTE, the sequence length N = 839. Multiple preamble sequences can be derived from one Zadoff-Chu sequence by cyclically shifting the sequence. Each

CP
684 s

RA sequence
1600 s

RA sequence

time

d) Format 3

Figure 1. LTE random access formats for FDD.

cell is assigned 64 preambles [14]. For small cells up to 1.5 km radii all 64 preambles can be derived from a single root sequence and are therefore orthogonal to each other. In larger cells not all preambles can be derived from a single root sequence and multiple root sequences must be allocated to a cell. Preambles derived from different root sequences are not orthogonal to each other, but the cross-correlation is low. High mobile velocities relative to the base station cause additional correlation peaks, which lead to ambiguous timing determination. In order to cope with this problem in LTE highspeed mode root sequences are defined, for which certain cyclic shift values are disabled so that transmitted preamble and round trip time can uniquely be identified. 2) Random Access Formats The transmitted preambles travel along different paths to the receiver, causing a delay spread of the received preamble. By adopting a cyclic prefix (the last part of the preamble is copied and prefixed the preamble before transmission), the receiver can suppress this spread. Furthermore, random access coverage is related to the maximum transmission power of the mobile. For large cells, some mobiles are unable to provide the receiver with sufficient received energy for correct preamble detection due to this limit. An alternative means to increase the received energy at the receiver is to transmit for a longer time. Therefore, some formats feature a repetition of the preamble. Figure 1 illustrates the four random access formats for LTE frequency division duplex (FDD), featuring both short and long cyclic prefix corresponding to capabilities to handle narrow and wide delay spreads, and single and repeated random access preambles to enable coverage in cells with different sizes [14]. Format 0 fits within 1ms (one subframe), format 1 and 2 fits within 2ms, while format 3 fits within 3ms. 3) Random Access Opportunities

30

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

frequency PRACH 1 RB PUSCH PUCCH 6 RB time 1 ms 20 ms

Correlation zone Correlation

Detection threshold

preamble Noise and interference time TCS Roundtrip time estimate

Figure 2. Example of random access opportunities. The opportunities repeats every 20 ms.

Figure 4. Random access preamble correlation detector, and round trip time estimation for cyclic shift length TCS.

resolved. The procedure is outlined in Figure 3, and the steps are further described below. 1) Random Access Preamble The mobile selects a preamble and an opportunity at random, and determines the format based on the broadcasted system information. In addition, the mobile determines the preamble transmission power by estimating the downlink path loss PL from the downlink reference signal (pilot signal) and using the broadcasted parameters P0_RACH (the desired received power), RACH (the power ramping step) and Preamble (the preamble-based offset equal to 0 dB for formats with a single preamble, i.e., formats 0 and 1 and equal to 3dB for formats with duplicate preambles, i.e., format 2 and 3). The mobile also monitors the preamble transmission attempt number m. For the initial preamble transmission, m=1, and the preamble transmission power is set according to
PRACH = min{Pmax , P0 _ RACH PL + ( m 1) RACH + Preamble }

Figure 3. Contention-based random access procedure for LTE with four steps.

In the classical random access scheme Slotted Aloha, access attempts are restricted to slots to avoid partial overlap between users. Similarly, in LTE the reserved time-frequency resources for random access the random access opportunities are slotted, and the mobile selects an opportunity at random among the available opportunities, see Figure 2. The considered resources for the physical random access channel, PRACH, can also be allocated to the physical uplink shared channel (PUSCH) used for scheduled uplink data transmission. The latter channel can also carry uplink control information as an alternative to the less flexible and capable physical uplink control channel (PUCCH). Therefore, the resource allocation needs to consider the balance between the PRACH and PUSCH demands. The plausible random access opportunities [14] dictate both the opportunity period and the timing, for example enabling nonoverlapping opportunities for three cells at the same site. Furthermore, the opportunity selection is also related to the random access format, since the opportunities needs to be sparse enough to fit the length of the selected format (1, 2 or 3 ms), while avoiding preamble overlaps. B. Contention-Based Random Access The contention-based random access procedure can be applied to all random access events. It is possible that at least two mobiles select the same resources (preamble and opportunity) for random access, and therefore the contention situation needs to be

(1)

Finally, the selected preamble is sent with the determined power level and format in the selected opportunity. 2) Random Access Response The base station correlates the received signal in each random access opportunity with all possible preamble sequences. Figure 4 illustrates the detector and the corresponding round trip time estimation subject to noise and interference. Upon detection of a preamble in an opportunity, the base station signals timing adjustment information, and an uplink resource allocation, and all mobiles that used the specific preamble in the specific random access opportunity considers this information. If no response is obtained within a configured time window, the mobile increases the preamble transmission attempt number m and returns to step 1) unless the max number of attempts has been reached. 3) Scheduled Transmission Using the allocated uplink resource, the mobile transmits an identity that uniquely identifies the mobile in the base station. 4) Contention Resolution During step 3 of the random access procedure, several mobiles that have sent the same preamble may respond. The base

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

31

station chooses one of the mobile identities and responds with the identity of the selected mobile, and only this mobile acknowledge the reception of the contention resolution. C. Non-Contention-Based Random Access This procedure is applicable when the base station can signal a reserved random access preamble to the mobile, i.e. at handover and uplink synch failure. In this case, all necessary information (essentially timing) is acquired at the random access response. III. REQUIREMENTS AND MODELING

The contention ratio is defined as,

na ( k ) , nd ( k ) > 0 1 CR (k ) = nd (k ) nd ( k ) = 0 0,

This section addresses both key performance metrics and their observation, and well as more general radio network simulation modeling. A. Performance Requirements and Observations The main objective with the random access procedure is to provide prompt and reliable access. Therefore, the access probability is of interest, i.e. the probability that a mobile acquires access upon an attempt. It is likely that an operator or vendor would like to specify the access probability APm at attempt m, i.e. the probability that the UE has access after attempt m (1mM), and then use autonomous algorithms to adjust the random access related parameters accordingly. For example, AP1 = 0.8 and AP3 = 0.99. Furthermore, it is instructive to express the access probability APm as a function of the detection miss probability and the contention probability. The detection miss probability DMPm at attempt m is defined as the probability of a preamble, transmitted at attempt m, not being detected at the base station. Moreover, the contention probability CP is defined as the probability that a UE is not granted access due to a preamble collision, conditional that the preamble of the UE is detected. The access probability at attempt m can therefore be expressed as

Finally, the access ratio is obtained via DMR and CR and (2). The number of detected preambles nd and number of mobiles that are granted access na are directly measurable at the base station and it is therefore tractable to estimate CR. However, it is not possible to measure ns at the base station unless this is reported by the mobiles. An undetected preamble is simply a correlation peak below the detection threshold (see Figure 4), which is classified as noise at the base station detector. Henceforth, we assume that mobiles report the number of attempts needed to obtain access once the mobile is granted access to the network (see the PREAMBLE_TRANSMISSION_COUNTER in [15]). These reports from the mobile enable the derivation of DMRm. Note that this particular mobile report is not yet standardized for LTE at the moment of writing. B. Radio Network and Random Access Modeling The simulation work in this paper is based on a semi-static simulator with random drops of mobiles without mobility modeling, but with time correlations (e.g. a failed random access attempt at one time instant will result in a retransmission by the same mobile at later time). It models the random access procedure in a multi-cell scenario with interfering uplink data traffic. The network is deployed in a hexagonal layout of 7 sites each 3-sectored and wrap-around propagation. The path loss predictions are adopted from [18] and the antenna models from [19]. The number of created mobiles at each drop that initiate the random access procedure follows a Poisson process with the mean arrival intensity LoadRACH (number of mobiles/second/cell), and they are uniformly distributed over the simulated area. Uplink data traffic is modeled by the PUSCH load (denoted LoadPUSCH), defined as the fraction of the frequency band that is used for PUSCH during a sumframe of 1ms (c.f. Fig. 4.). A cell with no random access opportunity at a specific subframe randomizes (depending on the PUSCH load) whether uplink data is scheduled in the frequency band where random access is configured. If uplink data is scheduled then a PUSCH user is randomized in the cell. Moreover, PUSCH power control is based on [16] and simplified to open-loop power control,

APm = 1 (DMPi + (1 DMPi ) CP ) .

i =1

(2)

The corresponding observables are denoted access ratio, detection miss ratio, and contention ratio. Assume that random access procedure data is collected over time intervals of length T, and let n(k) denote the counter value gathered over the time interval [(k-1)T,kT]. In particular, denote the number of sent preambles by ns(k), number of detected preambles by nd(k), and number of mobiles that have successful random access by na(k). An additional subscript m may be used to denote a particular attempt number. For example, ns,m(k) gives the number of sent preambles for attempt number m during the time interval [(k1)T,kT]. Hence, the preamble detection miss ratio for attempt m is given by,

PPUSCH = min{Pmax , P0 _ PUSCH + PL} dBm .

where P0_PUSCH is the desired target received power, PL is the path loss estimated by the UE based on the downlink reference signal, and Pmax is the maximum transmission power. Random access mobiles select a preamble (randomly) and an opportunity (typically the next available), and transmit at a power given by (1). However, the path loss estimate PL is measured for the downlink and not the uplink, and therefore additive white

n d ,m ( k ) , n s ,m ( k ) > 0 1 DMRm (k ) = ns ,m (k ) n s ,m ( k ) = 0 0,

32

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

Gaussian measurement noise with standard deviation EE is considered. The received preambles are processed and the signal to interference and noise ratio SINRp,c of each preamble p received at cell c is computed according to,
DMR

Attempt Nr 1 1

0.5

SINRp , c =

Pp g p ,c I PUSCH , c + I RACH ,c + N

0.2

0.4 0.6 PUSCH Load

0.8

where Pp is the transmission power of the mobile transmitting preamble p, gp,c is the path gain from the mobile to the based station of cell c, N is the thermal noise power over the random access frequency band, IPUSCH,c is the received interference power from PUSCH at cell c from users in other cells, and IRACH,c is the received interference power from random access preambles transmitted by other mobiles in the same cell (typically zero if all preambles originates from the same root sequence). The SINRp,c is then mapped to a preamble detection probability (see [21]). If several mobiles transmit the same preamble in the same opportunity in a cell, then contention resolution is carried out by randomly choosing a preamble (i.e. mobile) among the detected preambles. Table I summarizes central parameter values used in the simulations.
TABLE I. Parameter User distribution Site to site distance Antenna Tilt PMAX P0_PUSCH N SIMULATION PARAMETERS Value Uniform 500 m (5000m Section IV.E) 8 degrees 23 dBm W -89 dBm -109 dBm W 3 dB L = 128.1+37.6log10(d), d [km] 8 dB standard deviation DEFAULT EXPERIMENT SETUP Default Value 0.5 250 preambles/cell/s 0 5 ms -120 dBW 2 dB 8 180 s

(a)

0.02

0.015 CR

0.01 -120dBW -130dBW -140dBW

-150dBW

0.005

0.2

0.4 0.6 PUSCH Load

0.8

(b)
Figure 5. DMR and CR as a function of LoadPUSCH and P0_RACH.

EE

Path loss L Log-normal shadowing TABLE II. Parameter LoadPUCSH LoadRACH RACH Format RACH Opp interval P0_RACH

Note that the standard value for LoadRACH may seem too high. Since at the time of the writing LTE has not been deployed in large scale and typical loads are not yet available, we assume a wide range of RACH loads in the simulations. The default RACH load has been selected such that CR = 0.01 for one random access opportunity per 5 ms [14]. Also a higher RACH load enables shorter simulations times since more data is gathered compared to a lower RACH load. A. Effects of Varying PUSCH Load The goal of this experiment is to study the effects of P0_RACH and LoadPUSCH on DMR and CR. Recall that P0_RACH dictates the received signal power and LoadPUSCH determines the interference on RACH. The parameters are altered according to LoadPUSCH = {0.0,0.2,,1.0} and P0_RACH = [-150,-120] dBW in steps of 10 dBW. As shown in Fig. 5(a) the DMR of the first attempt (DMR1) increases with increasing LoadPUSCH and decreasing P0_RACH. The DMR of attempts 2-8 show similar behavior. Fig. 5(a) indicates that some P0_RACH values result in very low DMR and robustness to varying LoadPUSCH and interference. Consequently it seems that setting P0_RACH to, e.g., -130 dBW, will give a satisfactory RACH performance. However, these results hold only for the deployment used and the models and assumptions of, e.g., propagation, PUSCH and RACH. There may be cases where a P0_RACH lower (or higher) than -130 dBW should be used depending on prevailing conditions.

RACH

M Simulation Time

IV.

EXPERIMENTS

The experiment objectives are to illustrate the coupling between various tunable parameters, the performance of random access, and the interference caused by random access. Table II gives the standard parameters used in all experiments (if not otherwise stated).

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

33

Attempt Nr 1 0.8 0.6 DMR 0.4 0.2 0 -150

RACH=0 RACH=2 RACH=4 RACH=6

0.14 0.12 0.1 CR 0.08 0.06 0.04 0.02 0 0 200 400 600 RACH Load 800 RAOI = 20ms RAOI = 5ms RAOI = 2ms RAOI = 1ms

-145

-140

-135 P0_RACH

-130

-125

-120

Attempt Nr 3 0.8 0.6 DMR 0.4 0.2 0 -150

-145

-140

-135 P0_RACH

-130

-125

-120

Figure 7. Effect of RACH load and random access opportunity interval (RAOI) on CR.

Attempt Nr 5 0.8 0.6 DMR 0.4 0.2 0 -150

The conclusion of this experiment is that it is possible to control DMR by using P0_RACH and RACH. The parameter P0_RACH can be set according to the DMR requirements of the first attempt, whereas RACH can be tuned to satisfy DMR requirements for the other attempts. In some cases the latter may not be possible and in such circumstances P0_RACH must be adjusted as well.
-125 -120

-145

-140

-135 -130 P0_RACH

Figure 6. DMR as a function of P0_RACH and RACH

Fig. 5(b) shows that CR increases as LoadPUSCH increases and P0_RACH decreases. This is a result of an increasing number of preamble retransmissions (due detection misses) causing a higher contention probability. In conclusion, the PUSCH load and the induced PUSCH interference heavily affect DMR of all attempts. To counteract this, the power control parameter P0_RACH can be adjusted to an appropriate setting. Further, DMR and CR are coupled, meaning that an increase in DMR results in an increase in CR. B. Effects of Varying Power Control Parameters The goal of this experiment is to study the effects of P0_RACH and RACH on DMR and to establish whether a given performance specification in terms of DMR for each attempt number can be satisfied. The parameters are altered according to P0RACH = {120,-125,,-150} dBW and RACH = {0,2,4,6} dB. The results for DMR1, DMR3 and DMR5 are given in Fig. 6. In general, the DMR of all attempts decreases nonlinearly with increasing P0_RACH. As expected, for the first attempt the DMR does not vary over RACH. As such, the only way to control DMR for the first attempt is to set P0_RACH. For attempts greater than one, DMR varies over both P0_RACH and RACH. The amount by which DMR decreases when increasing RACH depends on the attempt number. This implies that for low attempt numbers there are limits for how much DMR can be altered by using RACH. As such, it may be necessary to alter P0_RACH to not only satisfy the first attempt, but also to satisfy attempt numbers greater than one.

C. Effects of Varying RACH Load and Configuration The goal of this experiment is to study the effects of RACH load and RACH configuration on CR. RACH load is altered according to LoadRACH = {100,300,,900} preambles/cell/s. RACH configuration corresponds to random access opportunity intervals (RAOI) of 20, 5, 2 , and 1 ms. Note that P0_RACH = -120 dBW, which results in the majority of the preambles to be detected at the first attempt. As expected CR increases with increasing LoadRACH and increasing random access opportunity period (determined by the RACH configuration), as shown in Fig. 7. The conclusion of this experiment is that it is possible to control CR by altering the configuration. D. Interference on PUSCH by Random Access Preambles The goal of this experiment is to study the interference on PUSCH generated by random access preamble transmissions. The idea is to show whether there is a benefit of adjusting P0_RACH in order to reduce the interference on PUSCH (compared to setting P0_RACH = -120 dBW). The generated interference is a function of the preamble transmission power and number preamble transmissions. For this reason we vary P0_RACH and the RACH load. The parameters are altered according to LoadRACH = {1,5,10,25,50,100,200,300} preambles/cell/s and P0_RACH = {-120,-130,-140,-150} dBW. Define the PUSCH noise rise as,

34

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

1 0.8 0.6 CDF 0.4 0.2 0

1 0.8 0.6

1 0.8 0.6 CDF 0.4 0.2 0

LoadRACH = 300 LoadRACH = 1

0 5 10 15 Noise Rise (dB) 20

CDF

0.4 0.2 0

10 15 Noise Rise (dB)

20

10 15 Noise Rise (dB)

20

(a)

(b)

(c)

Figure 8. Noise rise distributions as functions of P0_RACH and LoadRACH: (a) P0_RACH = -120 dBW (b) P0_RACH = -130 dBW (c) P0_RACH = -140 dBW.

Attempt Nr 1
Format=0 Format=2 Format=3

the performance requirements are satisfied). Note, the need of adjusting P0_RACH only is valid during excessive RACH load. E. Random Access Format Coverage Implication The goal of this experiment is to study the effect of RACH format on RACH coverage performance. Therefore, the P0_RACH and format are varied for a much sparser deployment with 5000 meters inter-site distance. From Fig. 9, it is evident that formats with repeated preambles improve the random access coverage. Furthermore, some mobiles in bad coverage spots will still have insufficient power to succeed with the random access, and raising the P0_RACH will not change this fact. A format with a repeated preamble will help some, but not all. V. SELF-OPTIMIZATION EXAMPLE

0.1 DMR

0.05

0 -150

-145

-140

-135 -130 P 0_RACH

-125

-120

Attempt Nr 2 0.08 0.06 DMR 0.04 0.02 0 -150

-145

-140

-135 -130 P0_RACH

-125

-120

Figure 9. DMR for first and second attempt and 5000 m inter-site distance.

NR =

I RACH ,tot + N RB N RB

where IRACH,tot denotes the received RACH inter-cell interference power (on PUSCH) and NRB -146 dBW is the noise power over one resource block (corresponding to a 180 kHz frequency band). The noise rise over all simulated cells is given in Fig. 8. The noise rise increases as a result of increasing LoadRACH and P0_RACH. The noise rise is substantial for some preamble transmissions when P0_RACH = -120 dBW and very small when P0_RACH = -140 dBW. Although, mobiles close to the base station may not suffer significantly from inter-cell RACH interference, mobiles at the cell edge may, and the result may be a decreased PUSCH coverage and/or performance. As such, there may be a benefit in decreasing the interference on PUSCH by lowering P0_RACH. The conclusion of this experiment is that RACH may cause interference on PUSCH for high P0_RACH and RACH load and this may be alleviated by appropriately setting P0_RACH (given

In order to exemplify how the information and models in this paper can form the basis for self-optimization, we illustrate that DMR1 can be controlled to meet a given performance specification by automatically adjusting P0_RACH. Recall from above that DMR is heavily affected by the PUSCH load. Therefore, LoadPUSCH is varied according to Fig. 10. Although the stepwise changes in LoadPUSCH may not be realistic, this gives the worst-case interference change on RACH and, allows us to study the performance of the controller under extreme conditions. The initial value of P0_RACH is -120 dBW. Detection miss probability for the first attempt should be 0.01. An integrating controller (I controller)
P0 _ RACH (k ) = P0 _ RACH (k 1) + K I (0.01 DMR1 (k ))

is used where KI is a tunable parameter, and sampling period is 1s. The results are given in Fig. 10, where the average over all cells is shown for LoadPUSCH, DMR1, and P0_RACH. We can see that the controller is capable of adjusting P0_RACH so that DMR1 tracks its target value (0.01). At time 45s and 80s, the PUSCH load increases significantly resulting in DMR1 overshoots. The overshoots cannot be avoided unless a mechanism that predicts the increase in LoadPUSCH is available. One conclusion is that using a simple I controller it is possible to control P0_RACH such

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops

35

LoadPUSCH

1 0.5 0

[2] [3]

50

100 Time (s)

150

[4]

0.05 DMR1 DMR Target 0

[5]

50

100 Time (s)

150

[6]

P 0RACH [dB]

-120

[7]
-130 0 50 100 Time (s) 150

[8]

Figure 10. Result of a self-tuning algorithm that autmatically sets P0_RACH based on observed DMR1.

that DMR1 satisfies a given performance specification in terms of a given target value. VI. CONCLUSION

[9] [10]

[11]

In this paper we have argued that there is an increasing need for self-organization in future wireless access networks. To meet these expectations there has been a significant effort carried out in academia, standardization bodies (e.g., 3GPP), and industry. One aspect that benefits from automation is RACH optimization, and the random access procedure is thoroughly described. We have studied the feasibility of RACH self-optimization in LTE by providing simulation results showing the impact of a set of key parameters on the RACH performance. Further an algorithm was presented, which tunes the RACH power control parameters such that the detection miss probability of transmitted preambles tracks given requirements. In order to automate the optimization of RACH power control parameters there is, however, a need for the UEs to report the number of sent preambles. These reports enable the derivation of preamble detection miss probability, which can be used for controlling the network access delay. REFERENCES
[1] NGMN, Use Cases related to Self Organising Network. Overall Description, ver. 1.53

[12] [13] [14] [15] [16] [17]

[18] [19]

[20] [21] [22] [23]

3GPP TR 36.902, E-UTRA Self-configuring and self-optimizing network use cases and solutions C. Prehofer and C. Bettstetter, Self-organization in communication networks: principles and design paradigms, IEEE Communications Magazine, July 2005. M. Amirijoo et al., Neighbor cell relation list and measured cell identity management in LTE, IEEE/IFIP Networking Operations and Management Symposium, 2008. D. Soldani and I. Ore, Self-optimizing neighbor cell lists for UTRA FDD networks using detected set reporting, IEEE Vehicular Technology Conference, 2007. G. Hampel, et al., The tradeoff between coverage and capacity in dynamic optimization of 3G cellular networks, IEEE Vehicular Technology Conference, 2003. K. Valkealahti, et al., WCDMA common pilot power control for load and coverage balancing, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, 2002. C. Lindemann, et al., Adaptive call admission control for QoS/revenue optimization in CDMA cellular networks, Journal of Wireless Networks, vol. 10, no. 4, July 2004, Springer. Socrates webpage, http://www.fp7-socrates.org/ J. Reig et al., Random Access Channel (RACH) Parameters Optimization in WCDMA Systems, IEEE Vehicular Technology Conference, Fall, 2004. S. Kim et al., Uplink Capacity Maximization based on Random Access Channel (RACH) Parameters in WCDMA, IEEE Vehicular Technology Conference, Spring, 2006. E. Dahlman, S. Parkvall, J. Skld, and P. Beming, 3G Evolution, second edition. 3GPP TS 36.300, E-UTRA and E-UTRAN Overall Description, Stage 2 3GPP TS 36.211, E-UTRA Physical Channels and Modulation 3GPP TS 36.321: E-UTRA Medium Acces Control (MAC) Protocol Specification 3GPP TS 36.213, E-UTRA Physical layer procedures B. M. Popovic. Generalized chirp-like polyphase sequences with optimum correlation properties. IEEE Transactions on Information Theory, Vol. 38, No. 4, Pages 1406-1409, July 1992. 3GPP TS 25.814, Physical layer aspect for evolved Universal Terrestrial Radio Access (UTRA). F. Gunnarsson et al., Downtilted Base Station Antennas A Simulation Model Proposal and Impact on HSPA and LTE Performance, IEEE Vehicular Technology Conference, Fall, 2008. M. Amirijoo et al., Towards Random Access Channel Self-Tuning in LTE, IEEE Vehicular Technology Conference, Spring, 2009. 3GPP R4-071951, PRACH Simulation Results, Ericsson 3GPP R1-080879, Power Control for PRACH, Ericsson J. Laiho, A. Wacker, T. Novosad, editors, Radio Network Planning and Optimization for UMTS, John Wiley & Sons, first edition, 2002.

36

2009 IFIP/IEEE Intl. Symposium on Integrated Network Management Workshops