Efficient Signaling for VoIP in OFDMA

Transcription

1 Efficient Signaling for VoIP in OFDMA Sean McBeath, Jack Smith, Doug Reed, Hao Bi 2, Danny Pinckley, Alfonso Rodriguez-Herrera, and Jim O Connor Motorola Fort Worth, Texas, 2 Libertyville, Illinois {sean.mcbeath, jack.smith, doug.reed, hao.bi, danny.pinckley, alfonso.rodriguez, j.o connor}@motorola.com Abstract Voice over internet protocol (VoIP) will be commonplace in future wireless standards. Due to the potentially large number of VoIP users in a wireless system, the overhead associated with controlling VoIP transmissions can hinder system performance, unless it is carefully managed. This paper outlines current efforts in B3G (beyond 3G) standards development to efficiently control VoIP transmissions by grouping VoIP users into scheduling groups, assigning the group a set of shared time-frequency resources, and using bitmap signaling to allocate resources. System level simulations are used to validate the signaling technique and show that the technique can efficiently support 93 users per megahertz. Potential further improvements are described. Keywords- orthogonal frequency division multiplexing; VoIP I. INTRODUCTION Efficiently supporting voice transmissions is a fundamental requirement for future wireless standards, such as those being developed by the Third Generation Partnership Project (3GPP) and Third Generation Partnership Project 2 (3GPP2) standardization bodies. Currently, 3GPP is developing the Evolved Universal Mobile Telecommunications Service Terrestrial Radio Access (E-UTRA) standard, while 3GPP2 is developing Revision C of the High Rate Packet Data (HRPD-C) standard. Both E-UTRA and HRPD-C rely on orthogonal frequency division multiple access (OFDMA) as the multiple access scheme for the forward link (FL). Packet based systems, such as HRPD-C and E-UTRA, can efficiently support voice over internet protocol (VoIP) traffic efficiently through the use of hybrid automatic repeat request (HARQ) and scheduling. The most straightforward technique for supporting VoIP for OFDMA is through the use of persistent resource assignments for each VoIP user. For persistent resource assignments, the access network (AN) assigns the access terminal (AT) a resource which persists at predefined intervals for a fixed period of time or until an known event occurs. The problem with persistent assignments is that the signaling overhead to allow the following can be prohibitive:. allowing other ATs in the system to use the resources when the VoIP AT sends an acknowledgement to the AN for one vocoder packet before the next vocoder packet arrives at the AN 2. allowing other ATs in the system to use the resources when the AN has determined a discontinuous transmission (DTX) state for a particular VoIP AT Therefore, in this paper, we describe a new signaling algorithm, which efficiently manages VoIP traffic. The scheme uses scheduling groups for VoIP traffic, where each scheduling group is made up of a set of VoIP users. Each scheduling group shares a common control channel and a common set of time-frequency resources. The set of timefrequency resources is divided into resource blocks, where each resource block is comprised of a number of orthogonal frequency division multiplexing (OFDM) symbols and a number of OFDM subcarriers. Further, bitmap signaling is used to efficiently indicate which users are receiving a resource block in each VoIP frame. Each vocoder frame (2 msec) is encoded and transmitted, using HARQ, in multiple non-contiguous VoIP frames. Each VoIP user determines its exact resource block(s) based on the information in the bitmap. Hence, the resource blocks for each AT can be changed from VoIP frame to VoIP frame. The scheme provides the following two forms of statistical multiplexing. First, statistical multiplexing is achieved among the group members. When the AN has determined a DTX state for a particular AT in a particular VoIP frame, no resource blocks are allocated to that AT. Second, statistical multiplexing is achieved between initial and subsequent HARQ transmissions. Once an AT acknowledges its VoIP packet, resource blocks are freed for other users in the group. In this way, the system can support more VoIP users than there are resource blocks. In this paper, we describe in detail the VoIP signaling technique developed as part of the 3GPP2 mobile broadband evolution and provide simulation results demonstrating the performance. The rest of the paper is organized as follows. In Section II, an overview of the OFDMA system is given. In Section III, the signaling technique is described. Section IV describes the VoIP packet formats. In Section V, simulation results are provided. Finally, the conclusions are given in Section VI. II. OFDMA SYSTEM OVERVIEW An OFDMA system is largely defined by its numerology and time domain structure. In this paper, the OFDMA system studied was based on the OFDMA numerology provided in Table I []. Referring to Table I, for a 5 MHz bandwidth, a 52 point Fast Fourier Transform (FFT) is used in conjunction with 32 guard subcarriers to provide 48 useful subcarriers. The subcarrier spacing is 9.6 khz. The subcarriers are grouped to form block resource channels (BRCH) and distributed resource channels (DRCH). A BRCH is a group of contiguous subcarriers that may hop within a /7/$ IEEE 2249

2 larger bandwidth, while a DRCH is a group of noncontiguous sub-carriers with some bandwidth. TABLE I OFDMA NUMEROLOGY Parameter Value Bandwidth 5 MHz FFT Size 52 Point Sampling Rate Msps Subcarrier Spacing 9.6 khz Cyclic Prefix 6.5 µsec Guard Subcarriers 32 Useful Subcarriers 48 Window Duration 3.26 µsec OFDM Symbol Duration 3.93 µsec Each OFDM symbol is 3.93 µsec. Eight OFDM symbols are concatenated to form a frame, which has a duration of.94 msec. Twenty-four frames and one preamble are concatenated to form a superframe, where the superframe duration is msec. Fig. illustrates the time domain structure of the studied OFDMA system. Vocoder Frame (2 msec) Superframe (22.94 msec) = VoIP Frame GroupID is used to control the entire group at once. For example, the AN can use the GroupID to assign or change the set of shared time-frequency resources which the group uses. Second, as part of assigning ATs to a group, each AT is assigned a unique position within the group. For example, AT 4 can be assigned the th position, AT 7 can be assigned the st position, AT 3 can be assigned the 2 nd position, and AT 29 can be assigned the 3 rd position. The position can be thought of as a unique identifier, which is valid only within the group. Third, once a group of ATs is established, the AN assigns the group a set of shared time-frequency resources and an ordering pattern indicative of the order in which the resources are allocated. In the time domain, the set of shared resources will be a group of VoIP frames comprising a VoIP interlace pattern. In the frequency domain, the shared resources will typically be a set of DRCHs, although a set of BRCHs could also be used. Fourth, each AT is assigned a VoIP interlace offset indicating in which VoIP frame of the VoIP interlace the transmission of the first HARQ subpacket will occur. Fig. 2 illustrates an example set of shared timefrequency resources and an example ordering pattern, while Fig. 3 illustrates the different VoIP interlace offset assignments that are possible within one VoIP interlace. = Preamble Ordering Pattern Fig.. Time Domain Structure Referring to Fig., one superframe is depicted, where two consecutive frames are defined to be a VoIP frame. A VoIP interlace is defined as a sequence of repeating VoIP frames, whereby initial and subsequent HARQ transmission occur in the same VoIP interlace. DRCH Index Time (frames) Fig. 2. Frequency Domain Resources III. BITMAP SIGNALING FOR VOIP To efficiently support VoIP traffic, a simple bitmap signaling structure is used to allocate resources in each VoIP frame. However, to support this structure, several preliminary steps are required. First, the AN assigns VoIP ATs to different scheduling groups, depending on channel quality indications (CQI) from the ATs. For example, there may be a QPSK group, a 6-QAM group, etc. Alternatively, all VoIP users in the system can be placed into the same group. To establish a group, the AN first assigns the group a unique identifier, namely a GroupID. Then, the AN assigns ATs to the group, using the unique identifiers of the ATs and the unique identifier for the group. For example, the AN can assign AT 4, AT 7, AT 3, and AT 29 to GroupID. The VoIP Interlace VoIP Interlace Offset Vocoder Frame (2 msec) Superframe (22.94 msec) 2 Fig. 3. Illustrative VoIP Interlace Offsets 225

3 Referring to Fig. 2, the set of shared time-frequency resources is comprised of 6 resource blocks. The term resource block will be used to indicate one frame by one DRCH. In particular, the set of shared time-frequency resources is one VoIP frame (two frames), where each frame contains 8 DRCHs. In this illustrative example, consider that each DRCH is 6 subcarriers. In Fig. 2, the y-axis is a logical representation of the frequency domain, since each DRCH is made up of 6 non-contiguous subcarriers. The shared time-frequency resources will repeat in each VoIP frame of the VoIP interlace. Referring to Fig. 3, three VoIP interlace offsets are defined, namely,, and 2. If an AT is assigned to VoIP interlace offset, the AN will transmit the first HARQ subpacket for that AT in the corresponding VoIP frame. Subsequent transmissions, if necessary, follow in the subsequent VoIP frames. This allows the AN to distribute the first HARQ transmissions in the time domain, which facilitates HARQ statistical multiplexing. Once a group of ATs is established and the set of shared time-frequency resources is defined, bitmap signaling is used to indicate which ATs are active in each VoIP frame. As an illustrative example, consider that a group consisting of 24 ATs has been established. Further, consider that the set of shared resources is 8 DRCHs by 2 frames, which is a total of 6 shared resource blocks as depicted in Fig. 2. In each VoIP frame, at least one, and possibly two bitmaps, are used to control which users are assigned which of the set of shared resources. The first bitmap is used to indicate the active ATs, while the second bitmap, if implemented, is used to indicate the number of resource blocks assigned to each AT. Each AT determines its allocation based on the allocations for all ATs with a smaller bitmap position in the first bitmap. Note that the bitmaps are sent at the beginning of each VoIP frame. Fig. 4 illustrates the two bitmaps. set of resources. This scenario may be advantageous when there are many more ATs in a group than blocks. As depicted in Fig. 4, the user with the first in the first bitmap corresponds to the first position in the second bitmap, the user with the second in the first bitmap corresponds to the second position in the second bitmap, etc. In this illustrative example, a in the second bitmap corresponds to an assignment of one block, while a in the second bitmap corresponds to an assignment of two blocks. To illustrate how ATs determine their assigned resources, Fig. 5 shows the allocations within the shared time-frequency resource blocks of Fig. 2 for the bitmap of Fig. 4. Referring to Fig. 2, Fig. 4, and Fig. 5, the first active wireless terminal, AT, is assigned one resource, and since it is the first AT allocated, it is allocated resource block of Fig. 2. The second active AT, AT 2, is assigned one resource block. AT 2 must sum the number of resources allocated to ATs with a smaller position in the first bitmap. In this case, AT 2 must determine that one resource block was previously assigned. Therefore, AT 2 is assigned resource block of Fig. 2. The third active AT, AT 4, is assigned one resource block. AT 4 must sum the number of resource blocks allocated to ATs with a smaller position in the first bitmap. In this case, AT 4 must determine that 2 resource blocks were previously assigned ( for AT and for AT 2 ). Therefore, AT 4 is assigned resource block 2 of Fig. 2. This process is repeated for all ATs. Fig. 5 shows the allocations for all active ATs in Fig AT 22 AT 23 AT 6 AT 9 AT 4 AT 6 Fig. 4. Example Fist and Second Bitmaps Referring to Fig. 4, the first bitmap is used to indicate active ATs. The bitmap locations correspond to the AT positions. For example, the AT assigned the th group position determines its assignment based on the th position in the first bitmap. Each AT with a in the first bitmap is active. The AT with the first is assigned the first M blocks, the AT with the second is assigned the second N blocks, etc, where M and N are the same if there is only the first bitmap, and M and N may be different if there are two bitmaps. If a second bitmap is used, the size of the second bitmap is equivalent to the number of ones in the first bitmap. Alternatively, the size of the second bitmap can be fixed and set to the number of blocks in the shared DRCH Index AT AT AT 8 AT 9 AT 5 AT 6 AT 4 AT 5 AT AT 2 Fig. 5. Example Assignemnts Time (frames) It is advantageous to define a mechanism to allow the AN to transmit two individually encoded packets to the same AT, for the cases when the AT does not successfully decode the packet after the third transmission. To accomplish this, the second bitmap can be configured to indicate that two independently encoded packets are being transmitted. When the second bitmap indicates that two resources are allocated, one of the resources is dedicated to 225

4 the first packet and the second resource is dedicated to the second packet. IV. VOIP PACKET FORMATS The simulations described in the next section used the cdma2 vocoder [2]. In particular, every 2 msec, the vocoder produces one of the following four transmission formats: full rate, half rate, quarter rate, and eighth rate. A full rate vocoder frame is 76 bits, a half rate vocoder frame is 8 bits, a quarter rate vocoder frame is 4 bits, and an eighth rate vocoder frame is 6 bits. The encoding structure is shown in Fig. 6. Referring to Fig. 6, RTP/UDP/IP (real-time transport protocol/user datagram protocol/internet protocol) overhead is modeled by adding 32 bits to the vocoder packet. Then, cyclic redundancy check (CRC) bits and an encoder tail allowance are added to the packet, after which the packet is encoded. It is known the convolutional codes outperform turbo codes for small block sizes, so a convolutional encoder is used for eighth and quarter rate packets and a turbo encoder is used for half and full rate packets. After encoding, the packet is interleaved and scrambled. Then, the Nth portion of the symbols are extracted from the packet for the Nth HARQ transmission. This process is referred to as subpacket symbol selection and repetition. Finally, the symbols are modulated and transmitted. Vocoder Packet 76 (Full Rate) 8 (Half Rate) 4 (Quarter Rate) 6 (Eighth Rate) Channel Interleaver + Long Code Mask Add 32-Bit RTP/UDP/IP Overhead Scrambler Add 6-Bit Packet CRC Subpacket Symbol Selection and Repetition Add Encoder Tail Allowance Fig. 6. VoIP Encoding Structure Encoder Modulate V. SYSTEM LEVEL SIMULATION RESULTS In order to study VoIP performance based on the bitmap signaling techniques, a series of system-level simulations were performed using a forward link OFDM system level simulator. The system level simulator complies with the cdma2 evaluation methodology [3] in most areas with the important simulation assumptions are given in Table II. TABLE II IMPORTANT SIMULATION ASSUMPTIONS Parameter Value Comments Number of Cells 9 2 rings, 3 sector system, Wrap-around universe ATs Per Sector 55 Per VoIP Interlace 7 deg (-3 db) Antenna Horizontal with 2 db front-toback Pattern ratio Antenna Orientation Propagation Model (AN Ant Ht=32m, AT=.5m) Log-Normal Shadowing degree horizontal azimuth is East (main lobe) log(d) db, d in meters Standard Deviation = 8.9 db No loss is assumed on the vertical azimuth. Modified Hata Urban Prop. (COST 23). Independently generate lognormal per mobile AN Correlation.5 AT Noise Figure. db Thermal Noise Density -74 dbm/hz Carrier Frequency 2 GHz AN Antenna Gain 7 db BS antenna 5 db with Cable Loss gain; 2 db cable loss AT Antenna Gain - dbi Other Losses db Fast Fading Model SCM Urban Macro 8 Delay Spread Model SCM Urban Macro 8 Max AN PA Power 8 Watts 5 MHz Bandwidth Site to Site distance 2. km Number of AT Antennas 2 Channel Estimation Non-Ideal Maximum C/I 7.8 db AT Speed Distribution 3 kmph (3% of ATs) kmph (3% of ATs) 3 kmph (2% of ATs) 2 kmph (% of ATs) kmph, fd=.5 Hz (% of ATs) Each AT is given a speed randomly selected from this distribution. The following four items are noteworthy aspects of the system level simulations, some of which differ from the cdma2 evaluation methodology. First, the spatial channel model (SCM) [4] was used. The SCM has a random power delay profile and power azimuth spectrum per drop. Therefore, an SCM draw can have a wide range of values for delay spread, angle spread at the AT, and angle spread at the AN. Second, the Equivalent SNR Method Based on Convex Metric (ECM) link-to-system mapping technique was used. The ECM has been shown to accurately predict the link level performance of randomly selected channel parameters like those defined by the SCM [5]-[6]. Third, the AT speed distribution listed in Table II was used. VoIP packets arrived according to a four state Markov model [2]. A technique known as eighth rate blanking was used, whereby only one out of every twelve consecutive eighth rate packets was actually transmitted over the air. All of the VoIP users in the system were placed into one group, and were controlled using the first and second bitmaps. All VoIP packets were modulated 2252

5 using QPSK. A jitter model was applied to arrival of the vocoder frames to account for queuing, contention, and serialization effects through an all-ip cellular core network [7]. The set of shared time-frequency resources used in the simulations was the entire 5 MHz of bandwidth and one VoIP interlace. The bandwidth was divided in 3 DRCHs, where each DRCH is made up of 6 subcarriers distributed over the entire 5 MHz. Four of the DRCHs were allocated for the transmitting the VoIP control message, which consisted of the first and second bitmap, with the second bitmap configured to indicate when two individually encoded packets were being transmitted. The simulations are intended to show the maximum amount of VoIP traffic that can be supported using a 5 MHz bandwidth. During simulation execution, each AT performs channel quality measurements over a VoIP frame using the transmitted pilots. These channel quality measurements are transmitted to the AN on the reverse channel quality indicator channel (R-CQICH) every 3 VoIP frames. Use of this CQI at the AN is delayed by two frames to properly account for feedback delay. Simulations were run for an increasing number of VoIP users until the outage criteria specified in [3] could no longer be achieved, which occurred at 55 ATs per VoIP interlace (93 ATs per MHz). In [8], the FL VoIP performance for HRPD Revision A was reported to be 4-45 ATs per.25 MHz (24-36 ATs per MHz). Therefore, the proposed technique represents a FL capacity improvement of 3-4 times that of previous systems. Figs. 7-9 illustrate the performance of the VoIP signaling technique at the operating point of 55 ATs per VoIP interlace. Fig. 7 shows the cumulative distribution function (CDF) of the time between when a vocoder packet arrived at the AN and when it was received at the AT. Prob(delay<=abscissa) delay (msec) Fig. 7. Delay CDF Referring to Fig. 7, the CDF indicates that 56% of the packets are delivered in less than 2 msec, which means that 56% of the packets are delivered before the next vocoder packet arrives. Further, notice that 99% of the packets are delivered in less than 42 msec. For a system where time-frequency resources are shared among several ATs, there can be unused system resources at each scheduling instance. In particular, due to eighth rate frame blanking and early HARQ termination, at any scheduling instance, both power and bandwidth can be unused. Fig. 8 shows the CDF of the amount of used bandwidth, while Fig. 9 shows the CDF of the amount of used power. Prob(Fraction of Bandwidth Used <= Abscissa) Prob(Fraction of Power Used <= Abscissa) Fraction of Bandwidth Used Fig. 8. Used Bandwidth CDF Fraction of Power Used Fig. 9. Used Power CDF Referring to Fig. 8, 5% of the time, 82% or less of the bandwidth is used. Referring to Fig. 9, 5% of the time, 84% or less of the power is used. These unused resources can be allocated to best effort data users to further improve system performance. The best effort data users can monitor the VoIP bitmap to determine which resources are available. 2253

6 VI. CONCLUSIONS In this paper, we introduced a new signaling mechanism for efficiently controlling VoIP transmissions. The signaling mechanism is advantageous, since it effectively takes advantage of the statistical multiplexing due to eighth rate blanking and early HARQ termination. Simulations were used to show that 93 ATs per megahertz can be supported, while still allowing 5-2% of the system resources to be allocated for best effort data users. In the development of this paper, a number of improvement mechanisms have been noted and are expected to improve capacity by 25% or more beyond what is shown in this paper. The following two improvement mechanisms are considered the most beneficial. First, the AN can define multiple VoIP groups wherein different modulation and coding requirements can be addressed. Second, the AN can use persistent assignments within the constraints of the grouping concept to reduce the control requirements. A detailed description of these improvement techniques and resulting performance will be part of a future paper. REFERENCES [] 3GPP2, C R4, Joint Proposal for 3GPP2 Physical Layer for FDD Spectra, July 26. [2] 3GPP2, C.S25, Markov Service Option (MSO) for cdma2 Spread Spectrum System, 2. [3] 3GPP2, C.R2-, cdma2 Evaluation Methodology, December 24. [4] 3GPP-3GPP2 SCM Ad Hoc Group, SCM-35: Spatial Channel Model Text Description, August 23. [5] A. Rodriguez, S. McBeath, D. Pinckley, D. Reed, Link-to-System Mapping Techniques Using a Spatial Channel Model, IEEE Veh. Technol. Conf., September 25. [6] S. McBeath, A. Rodriguez, D. Pinckley, and D. Reed, Applying the Convex Metric and the Spatial Channel Model for HRPD Rev-A, IEEE Wireless Communications and Networking Conference, April 26. [7] 3GPP2, C , Forward link VoIP packet delay jitter model, July 26. [8] W. Xiao, A. Ghosh, D. Schaeffer, L. Downing, Voice over IP (VoIP) over cellular: HRPD-A and HSDPA/HSUPA, IEEE Veh. Technol. Conf., September