`Amitava Ghosh, Rapeepat Ratasuk, Igor Filipovich, Jun Tan, Weimin Xiao
`Networks and Enterprise Business, Motorola
`1501 West Shure Drive, Arlington Heights, IL 60004, USA
`Abstract — Comprehensive long term evolution of the
`are signaled on the broadcast channel. In addition, these
`parameters may be configurable based on random access load.
`Universal Mobile Telecommunications System (UMTS)
`Note that data transmission may also be scheduled in the
`specifications is currently ongoing to provide significant
`random access regions at the discretion of the scheduler.
`improvement over the current release. Important goals for
`the evolved system include significantly improved system
`capacity and coverage, low latency, reduced operating
`costs, multi-antenna
`support,
`flexible
`bandwidth
`operations and seamless integration with existing systems.
`To ensure low latency, users must be able to establish a
`connection to the network quickly. This paper provides a
`preliminary design and procedure for the random access
`channel used to establish a connection when the mobile is
`not yet time-synchronized to the network in the uplink.
`I. INTRODUCTION
`With the emergence of packet-based mobile broadband
`systems such as 802.16e, it is evident that a comprehensive
`long term evolution (LTE) of UMTS is required to remain
`competitive. As a result, work has begun on Evolved UMTS
`Terrestrial Radio Access (E-UTRA) aimed at commercial
`deployment around the 2010 timeframe. Long term goals for
`the system include support for high peak data rates (100 Mbps
`downlink and 50 Mbps uplink), low latency (10ms round-trip
`delay, 100ms control plane delay), improved system capacity
`and coverage, reduced operating costs, multi-antenna support,
`efficient support for packet data
`transmission, flexible
`bandwidth operations (up to 20 MHz) and seamless integration
`with existing systems [1]. To reach these goals, a new design
`for the air interface has been adopted with Orthogonal
`Frequency Division Multiple Access (OFDMA)
`in
`the
`downlink and Single-Carrier Frequency Division Multiple
`Access (SC-FDMA) in the uplink.
`When the mobile is not time-synchronized to the base
`station in the uplink, it must use a contention-based random
`access channel to access the network. It may use this channel
`to request initial access, initiate handoff procedure, and
`transition from idle to connected state. To ensure low latency,
`random access procedure must be designed such that the
`control plane latency requirement of less than 100 ms is
`achieved. This paper provides a preliminary physical layer
`design and procedure for random access in E-UTRA.
`II. E-UTRA RANDOM ACCESS
`Figure 1 illustrates non-synchronized random access
`structure. In the figure, random access occupies a bandwidth
`of BWRA = 1.08 MHz and its length is a multiple of 1ms sub-
`frame. Multiple frequency regions may be defined within one
`access period in order to provide sufficient random access
`opportunities. In addition, this access period occurs at a fixed
`timing from the system broadcast frame. Although the timing
`is known, the number, location, and periodicity of the channels
`
`Figure 1. Non-synchronized random access.
`Within each random access region, multiple users may access
`the channel simultaneously by selecting different preambles.
`The preambles must exhibit good detection performance and
`robustness to interference as well as provide accurate timing
`estimation. This is because in E-UTRA uplink transmissions
`must be synchronized in order to prevent interference. In [8], it
`was shown that Ep/No (energy per sequence over noise) of
`approximately 16-18 dB is required to achieve 1% false alarm
`and missed detection probabilities. In addition, random access
`must be possible from the cell edge where the operating C/I
`may be very poor. As a result, to meet coverage requirement, it
`was determined that only the preamble (and no message) can
`be transmitted in the contention channel. From coverage
`perspective, it was determined that 64 different preambles can
`be supported per random access region.
`
`Figure 2. Non-synchronized random access burst.
`Figure 2 illustrates the baseline 1.0ms random access burst
`with cyclic prefix (TCP) of 100µs, preamble length of 800µs,
`and guard time (TGP) of 100µs. A cyclic prefix is added to aid
`in frequency domain processing in order to reduce detection
`complexity.
` A guard
`interval
`is required
`to prevent
`interference
`to other
`transmissions arising from
`timing
`misalignment. This timing misalignment between mobiles in
`the cell is dependent on the cell size and the 100µs guard
`interval corresponds
`to a supportable cell
`radius of
`approximately 14 km. However, in E-UTRA, random access
`
`1550-2252/$25.00 ©2007 IEEE
` 1041
`Authorized licensed use limited to: Univ of Calif Davis. Downloaded on December 05,2021 at 05:10:45 UTC from IEEE Xplore. Restrictions apply.
`
`APPLE 1017
`
`1
`
`
`
`possible contention (i.e. more than one mobiles selecting the
`same preamble sequence).
`In the third step, the mobile transmits its RRC connection
`request in the uplink using its temporary C-RNTI. Included in
`this message is the mobile identifier and whether this mobile
`has already been assigned a C-RNTI from its previous network
`access. This message is of dynamic size and H-ARQ can be
`used to ensure it is successfully received at the base station.
`The temporary C-RNTI will serve as its identity for contention
`resolution purposes which the base station would echo in the
`fourth message. This would serve as an early indication if a
`collision occurred during the previous transmission and allow
`the mobile to reinitiate random access procedure as soon as
`possible. Otherwise, the mobile will have to wait until the fifth
`message before contention is resolved. This may incur
`significant delay since the RRC connection request response
`has to come from the Access Gateway.
`B. Random Access Load and Overhead
`Non-synchronized random access will be used for initial
`access, handoffs to a non-synchronized cell, and scheduling
`requests. In general, random access load will be dominated by
`handoff requests. For random access load approximation, a
`traffic model representing busy hour traffic statistics was
`provided in [10]. In this model, two different traffic types
`were considered – real-time service such as VoIP and non real-
`time service such as web browsing. For real-time service, an
`average of 1 call per hour per user is assumed with the call
`duration of 90 seconds. For non real-time traffic, an average of
`2 calls per hour per user is assumed with duration of 300
`seconds. To estimate the number of handoffs, a mobile is
`assumed to change serving cell every 20 seconds. Using this
`traffic model, Figure 5 illustrates the probability of collision as
`a function of the number of camped mobiles within the cell for
`different random access overheads.
`Random Access Collision Probability (10 MHz System BW)
`
`2.0e-2
`
`0.6%
`
`1.2%
`
`2.4%
`
`3.6%
`
`4.8%
`
`0.6% random access overhead
`1.2% random access overhead
`2.4% random access overhead
`3.6% random access overhead
`4.8% random access overhead
`
`1.5e-2
`
`1.0e-2
`
`0.5e-2
`
`Probability of Collision
`
` 0.0
`100
`
`10000
`
`1000
`Number of Camped Mobiles
`Figure 5. Random access collision probability.
`From the figure, it is seen that amount of random access
`overhead required is naturally dependent on the target collision
`probability and cell load. For instance, at 5e-3 collision
`probability and 7000 camped mobiles, 3.6% of the system
`bandwidth is required to support random access. This
`
`must be designed to support large cells of size up to 100 km.
`In this case, repetition is used to extend the random access
`burst with appropriate adjustments to the cyclic prefix and
`guard period.
`
`Figure 3. Extended random access burst for large cells.
`For instance, to support a 25 km cell, an extended 2ms random
`access burst is deployed as shown in Figure 3 where the 800µs
`preamble is repeated twice with the cyclic prefix length and
`guard interval each extended to 200µs.
`III. RANDOM ACCESS PROCEDURE
`
`A. Access Procedure
`Figure 4 provides the random access procedure for initial
`access. Four different messages are exchanged as part of the
`random access procedure and contention resolution. They are
`(1) random access preamble, (2) random access response, (3)
`RRC connection request, and (4) RRC contention resolution.
`First, a mobile randomly selects a random channel. Next, it
`selects a random preamble from the set of available preambles.
`The mobile then determines the initial power setting using
`open-loop power control and transmits the preamble.
`
`Figure 4. Random access procedure.
`Upon reception of the random access preamble, the base
`station sends a response in a semi-synchronous manner in
`order to allow for some scheduling flexibility and load
`balancing. This response is sent using a combination of L1/L2
`control and downlink shared data channels. The L1/L2 control
`channel points to the location within the shared data channel
`where the actual random access response is contained. Note
`that multiple responses may be multiplexed into the shared
`data channel. Each random access response contains an uplink
`scheduling grant for RRC connection request transmission,
`timing advance information, and an assignment of a temporary
`C-RNTI (Cell Radio Network Temporary Identifier). Timing
`advance
`information
`is used by
`the mobile
`to
`time-
`synchronized its uplink transmission. Note that H-ARQ is not
`used for transmission of the random access response due to
`
` 1042
`Authorized licensed use limited to: Univ of Calif Davis. Downloaded on December 05,2021 at 05:10:45 UTC from IEEE Xplore. Restrictions apply.
`
`2
`
`
`
`translates to three random access channels for each 10ms radio
`frame. If the collision probability can be relaxed to 1e-2, then
`the overhead is correspondingly reduced to 2.4%. It should be
`noted that the traffic model in [10] is somewhat simplistic as
`only two traffic types are considered. In some situations the
`actual random access load may be significantly higher.
`Nonetheless, it may be inferred from the analysis that random
`access overhead of 3-6% should be sufficient to handle most
`deployment scenarios.
`C. Latency Analysis
`A key requirement of E-UTRA is to limit control plane
`latency to less than 100 ms. From Figure 4, it can be seen that
`random access failure due to for example poor radio conditions
`or missed detection can happen in each of the four messages.
`In addition, contention resolution is not performed until the
`fourth message. In [11], it was shown that a minimum of 12
`ms is required to complete the RRC contention resolution step
`and that a large percentage of users is able to complete the
`random access procedure in this time frame. However, 30-50
`ms access time may be required for a small percentage of the
`mobiles (less than 5%) due to contention. Nonetheless, from
`the analysis shown in [11] it may be concluded that control
`plane latency of less than 100 ms is feasible with this
`procedure.
`D. Simultaneous Data Transmission
`Since low latency is an important requirement, random
`access channels must occur with sufficient frequency (e.g.
`every 10 ms or so) which may result in high bandwidth
`overhead especially for system with smaller bandwidth. For
`10 MHz system, it is seen that random access overhead of 3-
`6% may be needed. In [8], it was shown that received
`sequence energy over noise of 16-18 dB is required to meet 1%
`false alarm and missed detection probabilities. For the 1ms
`preamble structure, this energy is spread over 800µs resulting
`in very low preamble SINR requirement per sub-carrier. Thus,
`it should be possible to transmit data simultaneously on some
`or all of the resource blocks used for non-synchronized random
`access. This is especially beneficial when systems are
`temporarily experiencing low random access loads (e.g. in the
`middle of the night). Naturally, the decision to schedule data
`users in those resource blocks lies with the base station and
`other factors such as possible interference should also be taken
`into consideration. In practice, the base station may restrict
`data transmission only to low QoS traffic such as best effort
`services and then for example using very low data rates to
`minimize possible interference. As a result, it is being
`proposed that data transmission may also be scheduled in the
`random access region at the discretion of the scheduler.
`E. Preamble Message Content
`Because random access load is expected to be small, it
`should be possible to convey some implicit information to the
`base station while still maintaining low overall random access
`collision probability. In this case, approximate knowledge of
`channel condition (e.g. good or bad channel) can substantially
`
`reduce the amount of resource required for the preamble
`response. This information may be used as a guide for
`downlink power control or link adaptation and possibly for
`uplink resource assignment. Without any channel knowledge,
`fixed downlink transmission power to reach the desired cell
`coverage must be used. This is substantial as the downlink C/I
`may be as low as -4 dB at the 95% cell coverage for systems
`with 1732 meter inter-site distance and 20 dB penetration loss
`[6]. At this operating point, the spectral efficiency is
`approximately 0.3 b/s/Hz [7]. Thus, to convey an exemplar
`40-bit random access response will require approximately 130
`sub-carriers, which is a substantial overhead.
`Table 1. Average control channel power saving.
`No of bits used to convey
`Average Power Saving
`channel condition
`(dB)
`0
`0.0
`1
`4.5
`2
`6.7
`Table 1 lists potential saving in downlink control channel
`power allocation if some rough channel quality information is
`available assuming uniform distribution of users. From the
`table, it is seen that substantial control channel overhead
`saving (in power or bandwidth) can be achieved with
`knowledge of channel quality information. Although the
`addition of this information will correspondingly reduce the
`number of random preambles,
`the effect on collision
`probability can be minimized through careful partitioning of
`the user space.
`F. Dedicated Preambles
`In general, UE randomly selects a preamble from the
`available set for
`transmission with possible contention.
`However, several proposals have been made to reserve a subset
`of preambles for dedicated purposes such as handover, uplink
`synchronization, and feedback. Depending on the load and
`access pattern, this option may be more efficient than using
`random preambles. In this case, preambles are reserved and
`allocated to UEs so that they may be use in a contention-free
`manner. For example, during handover the target base station
`can allocate a specific preamble to a user making the transition
`so that the handover process can be performed in a contention-
`free manner. For some applications, it may be sufficient to
`simply reserve the preambles without having to allocate them
`to users. For instance, indicative feedback regarding broadcast
`service quality can be provided using just one common
`preamble because user-specific information is not required (i.e.
`the networks only need to know that some users are
`experiencing poor radio conditions, but not which users). In
`this case, responses from multiple users are transmitted using a
`common preamble.
` Some energy aggregation may be
`performed to further gauge the approximately number of
`responses. Naturally, with the use of dedicated preambles,
`capacity on the non-synchronized random access channel is
`correspondingly reduced and more time-frequency regions
`may be needed to maintain low collision probability for other
`random access users.
`
` 1043
`Authorized licensed use limited to: Univ of Calif Davis. Downloaded on December 05,2021 at 05:10:45 UTC from IEEE Xplore. Restrictions apply.
`
`3
`
`
`
`zero for all delay not equal to an integer multiple of sm. As a
`result, the resulting sequence has a zero correlation zone of
`length sm-1.
`Since in general multiple ZC-ZCZ or GCL-ZCZ sequence
`sets may be needed to generate the required number of
`preambles, sequence performance will be affected by the cross-
`correlation performance between groups. For the ZC-ZCZ,
`the periodic cross-correlation between any two sequences is
`constant. For the GCL-ZCZ, the cross-correlation depends on
`selected indices. Figure 6 illustrates the normalized average
`cross-correlation at zero delay between the sequences when
`more than one Zadoff-Chu carrier sequences are used. From
`the figure, it is seen that when the number of groups is small,
`Zadoff-Chu
`and GCL have
`similar
`cross-correlation
`performance. However, as the number of groups increases, the
`cross-correlation of the GCL sequence is significantly worse
`than that of the Zadoff-Chu.
`
`Sequence Cross-correlation - ZC-ZCZ(N=509) and GCL-ZCZ(N=512)
`
`ZC-ZCZ
`GCL-ZCZ
`
`0.25
`
`0.2
`
`0.15
`
`0.1
`
`0.05
`
`Normalized Avg Cross-Correlation
`
`0
`
`0
`
`5
`
`25
`20
`15
`10
`No of Zadoff-Chu Carrier Sequence
`
`30
`
`35
`
`Figure 6. Normalized average cross-correlation between ZC-
`ZCZ and GCL-ZCZ.
`the analysis, sequence comparison may be
`Based on
`summarized as follows –
`• Both Zadoff-Chu and GCL can be designed to have the
`same number of zero-correlation zones based on cell
`range. In general, multiple root sequences may be
`needed within each cell which will
`introduce
`interference between sequences from different roots.
`• The number of available GCL root sequences is less
`than that for Zadoff-Chu. Therefore, preamble planning
`with GCL will be more complicated.
`• As the number the root sequences deployed within a cell
`increases, GCL cross-correlation between groups
`worsen. On the other hand, it remains the same for
`Zadoff-Chu.
`As a result, Zadoff-Chu sequence with cyclic shift was selected
`as the preamble for E-UTRA. For the baseline preamble
`length of 800µs, this corresponds to a sequence of length 863
`samples.
`B. Sequence Planning
`An important aspect of system design is sequence
`planning for random access. Based on the chosen sequence
`design, 862 different Zadoff-Chu sequences are available. For
`
`IV. PHYSICAL LAYER DESIGN
`A. Preamble Sequence Design
`Naturally, preamble waveforms for random access should have
`good detection probability while maintaining low false alarm
`rate, allow accurate timing estimation, and low power de-
`rating. Two promising sequences were analyzed - Zadoff-Chu
`with Zero Correlation Zone (ZC-ZCZ) [3] and Generalized
`Chirp-Like with Zero Correlation Zone (GCL-ZCZ) [4]. A
`sequence set with zero correlation zone property has zero
`periodic cross-correlation for a contiguous set of delays. Both
`sequences belong to a family of Constant Amplitude Zero
`Auto-Correlation
`(CAZAC)
`sequences.
` The constant
`amplitude results in low peak-to-average power ratio in the
`transmitter. This is especially important in the uplink where
`peak-to-average power ratio must be kept low due to power
`amplifier limitations.
`The Zadoff-Chu sequence of length N is given by the
`expression
`
`21
`
`ππ
`
`21
`
`−
`
`j
`
`−
`
`j
`
`(cid:176)(cid:175)(cid:176)(cid:174)(cid:173)
`
`g
`
`p
`
`
`
`)(n
`
`=
`
`2
`e
`N
`M
`2
`e
`N
`
`odd is
`M
`where p, the sequence index, is relatively prime to N (i.e. the
`only common divisor for p and N is 1). For a fixed p, the
`Zadoff-Chu sequence has
`ideal periodic auto-correlation
`property (i.e. the periodic auto-correlation is zero for all time
`shift other than zero). For different p, Zadoff-Chu sequences
`are not orthogonal, but exhibit low constant cross-correlation
` regardless of time shift. If the sequence length N is
`of
`N/1
`selected as a prime number, there are N – 1 different available
`sequences. The zero-correlation zone for the Zadoff-Chu
`sequence is generated using cyclic shift version of the base
`carrier sequence. Note that each zero-correlation zone must be
`large enough
`to accommodate
`the maximum
`timing
`misalignment between mobiles in the cell, which is dependent
`on the cell size. Thus, the number of zero correlation zones
`that can be generated per each sequence index p is based on the
`sequence length N and the cell size. The maximum available
`number of sequences available in the system is then (N-1)×L
`where L is the number of zones per sequence index.
`The Generalized Chirp-Like (GCL) sequence of length N
`is defined as
`g
` ()( nbn
` ), nm
`N
`
` )(nc
`
`
`
`
`mod
`,...,1,0
`1
`=
`=
`−
`p
`with the sequence length N satisfying the relationship N=sm2
`where s and m are positive integers. Note that length
`constraint of the GCL sequence can be relaxed to N=tm where
`t and m are positive integers [5]. The carrier sequence gp(n) is
`the Zadoff-Chu sequence of length N. As before, the sequence
`index p must be a relative prime to N. Due to the restriction on
`the sequence length, the number of available GCL groups is
`generally less than that for Zadoff-Chu. To provide a set of
`orthogonal GCL sequences, a common Zadoff-Chu sequence is
`modulated by m different orthogonal sequences {bi(k)},
`i=0,…,m-1. An example of orthogonal modulation is the
`Hadamard sequence. The periodic cross-correlation between
`any two GCL sequence with the same Zadoff-Chu carrier is
`
`2
`
`pn
`
`
`
`(npn
`)1
`+
`
`
`
` even is
`
` 1044
`Authorized licensed use limited to: Univ of Calif Davis. Downloaded on December 05,2021 at 05:10:45 UTC from IEEE Xplore. Restrictions apply.
`
`4
`
`
`
`transmission then occurs using the selected random access
`channel, preamble sequence, and preamble transmission power.
`4. If no response corresponding to the transmitted preamble
`sequence is detected then another random access channel and
`preamble are randomly selected. If the maximum transmission
`power and the maximum number of retransmissions have not
`been reached, then preamble retransmission occurs. Otherwise
`the L1 status is passed to the higher layers and the physical
`random access procedure is terminated.
`5. If a response corresponding to the transmitted preamble
`sequence is detected, then the L1 status is passed to the higher
`layers and the procedure is terminated.
`V. CONCLUSIONS
`In this paper, a preliminary look at the design of the
`random access channel and procedure for E-UTRA is provided.
`From the analysis, it is apparent that the E-UTRA requirements
`of low latency and overhead can be achieved. Specifically, the
`design provides a flexible approach where access regions can
`be added as random access load grows and where sequence
`planning can be tailored based on supported cell size.
`Random access overhead of 3-6% is seen to be sufficient for
`most system deployment scenarios. In addition, sequence
`planning should not be a concern due to the large number of
`available Zadoff-Chu root sequences. Finally, control plane
`latency of less than 100 ms is feasible with this random access
`procedure.
`
`REFERENCES
`[1] 3GPP TR 25.913, Requirements for Evolved UTRA (E-UTRA)
`and Evolved UTRAN (E-UTRAN), v.7.3.0, March 2006.
`[2] 3GPP TR 25.814, Physical Layer Aspects for Evolved UTRA,
`v.2.0.0, June 2006.
`[3] D. C. Chu, “Polyphase codes with good periodic correlation
`properties,” IEEE Trans. Inform. Theory, vol. 18, pp531-532,
`July 1972.
`[4] B. M. Popovic, “Generalized Chirp-Like Polyphase Sequences
`with Optimum Correlation Properties,” IEEE Trans Inform.
`Theory, vol. 38, no. 4, pp1406-1490, July 1992.
`[5] R1-062128, “Generalized Construction of ZCZ-GCL Random
`Access Preambles,” Huawei, RAN1#46, Tallinn, Estonia, Aug.
`2006.
`[6] R1-060009, “E-UTRA Downlink Control Channel Design and
`Performance and TP,” Motorola, RAN1 LTE Ad Hoc, Helsinki,
`Finland, Jan. 2006.
`[7] R1-061164, “Downlink Control Channel Modulation and
`Coding,” Motorola, RAN1#45, Shanghai, China, May 2006.
`[8] R1-061166, “Random Access Payload Size,” Motorola,
`RAN1#45, Shanghai, May 2006.
`Initial C-RNTI
`[9] R2-062477, “Contention Resolution and
`Allocation,” Motorola, RAN2#54, Tallinn, Estonia, Aug. 2006.
`[10] R2-062160, “RACH Contention and Retry Cases,” NTT
`DoCoMo, RAN2#54, Tallinn, Estonia, Aug. 2006.
`[11] R2-070268, “RACH Budget and Delay Analysis,” NTT
`DoCoMo, RAN2#54, Tallinn, Estonia, Aug. 2006.
`Note – 3GPP documents may be downloaded from ftp://ftp.3gpp.org
`
`each sequence, a number of possible cyclic shifts are available
`based on the cell radius to be supported. However, the number
`of possible cyclic shifts should be restricted to a power of two
`so that 64 preambles can be exactly constructed from the
`appropriate set of Zadoff-Chu root sequences. Naturally, only
`one value of cyclic shift is used within a cell based on the
`supported cell size. Figure 7 illustrates the number of root
`Zadoff-Chu sequences required to provide 64 preambles based
`on the maximum cell radius.
`70
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`Maximum Cell Radius (km)
`
`0
`
`1
`
`2
`
`32
`
`64
`
`16
`8
`4
`Number of Root Zadoff-Chu Sequence
`Figure 7. Number of root Zadoff-Chu sequences required.
`From the figure, it is seen that in most cases sequence planning
`should not be a concern due to the large number of available
`Zadoff-Chu root sequences. For instance, even for 30 km
`cells, a sequence reuse factor of 53 is provided. Additional
`frequency reuse is possible by deploying random access
`regions in different frequency regions among cells.
`C. Physical Layer Access Procedure
`From the physical layer perspective, the L1 random access
`procedure encompasses successful transmission of the random
`access preamble and response. Open-loop power control with
`ramping is used in the preamble transmission in order to
`reduce interference. The following steps are required for the
`L1 random access procedure:
`1. Prior to initiation of the non-synchronized physical random-
`access procedure, Layer 1 shall receive
`the following
`information from the higher layers - random access channel
`parameters, preamble format for the cell, number of root
`Zadoff-Chu sequences and sequence indices, power ramping
`step
`size, and
`the maximum number of preamble
`retransmissions.
`2. A random access channel is randomly selected from the
`available non-synchronized random access channels.
` A
`preamble sequence is then randomly selected from the
`available preamble set based on the implicit message to be
`transmitted. The random function shall be such that each of
`the allowed selections is chosen with equal probability.
`3. The initial preamble transmission power level (which is set
`by the RRC layer) is determined using an open loop power
`control procedure. The transmission counter is set to the
`maximum number of preamble retransmissions. Preamble
`
` 1045
`Authorized licensed use limited to: Univ of Calif Davis. Downloaded on December 05,2021 at 05:10:45 UTC from IEEE Xplore. Restrictions apply.
`
`5
`
`