WiMax MBS Power Management, Channel Receiving and Switching Delay Analysis

Thawatt Gopal (Huawei Technologies USA) 10180 Telesis Court, Suite 365, San Diego, CA 92121, USA [email protected]

Abstract In Wimax MBS (Multicast-Broadcast System), the power savings and channel switching delay are two critical performance metrics that affects the viewing time capacity and quality of experience of the end-user. The amount of air-interface resources allocated for MBS, power savings targeted, channel receiving/switching delay target, multimedia stream data rate are all inter-related. This paper analyzes the interactions of these parameters for Wimax MBS and quantifies the impact of key parameters on the desired power savings ratio and channel switching delay. Keywords MBS, Wimax, MBS Power Management, MBS Channel Switching
I. INTRODUCTION

many MBS streams can be supported for a given power savings target. The switching delay is an important metric that is frequently compared for MBS type of services related to watching broadcast video. The longer the time between data bursts for the same MBS stream, the better the power savings benefit whilst the channel switching performance degrades.
II. POWER MANAGEMENT

There are several key performance metrics for MBS (Multicast-Broadcast System) type of services. This includes the areas of power management, channel switching delays, RF coverage in a SFN (Single Frequency Network), application layer throughput and packet error rate. This paper covers the areas of power management and channel switching delay. For power management, we make use of models originally developed for the DVB-H system and apply them to a Wimax MBS system. There are certain similarities between DVB-H and Wimax MBS namely both uses OFDMA based airinterfaces and the concept of time-slicing introduced in DVBH systems generally applies to Wimax MBS also. Timeslicing was introduced in DVB-H to prolong the terminal battery life whereby if the multimedia stream rate was Rc, then the air-interface burst bitrate is MRc where M is a function of the supported air-interface capacity and targeted power savings percentage. This allows the mobile station to save power while not receiving MBS data that it has not subscribed or wants to receive and allows the multiplexing of several multimedia streams that can be decoded independently in time. Several differences to note between Wimax MBS and DVB-H are that Wimax supports simultaneous unicast and MBS traffic in the downlink frame. It is also possible to multiplex several multimedia streams in the same MBS Zone in a Wimax frame. The MBS Zone is the portion of the Wimax frame that can be allocated for MBS data. The amount of bandwidth allocated determines how

In terms of power management, a model incorporating the ON, OFF and Synchronization time (Tsync) provides target values of the power savings percentage. The Synchronization time is the amount of time required for the mobile device to wake-up from sleep and get ready to start demodulation and decoding of the MBS data stream. Contributions to [1] from mobile manufacturers suggest that Tsync should be assumed in the range of 200 to 250ms with power savings target of 90% but here we investigate a range of power savings target. For a range of existing handheld device battery capacity, we provide several points of references. The HP iPAQ 210, has a relatively large battery capacity of 2200 mAh rating with features targeted at mobile computing and according to tests done in [3] for viewing streaming video it has a battery life of 3 hours and 28 minutes. Another example is the Nokia E61 (targeted at the mobile computing application) has a battery capacity of 1500 mAh while a simpler device like the Motorola RAZR v3 has a battery life of 710 mAh. Another point of reference from Media Flo [4] indicates a target video viewing time of 3.9 hours for video rates of 360 kbps with a 850 mAh battery capacity assuming some form of power management/bursting technique is used while the test results for the HP iPAQ does not consider such technique. In DVB-H, an OFDMA frame is used entirely for broadcast services while in Wimax MBS the MBS burst (specified by number of OFDMA symbols) occupies a portion of the OFDMA frame called the MBS Zone. For a given power savings percentage target, given average stream bitrate, we want to find out what is the required burst bitrate. The following diagram and equations shows the power saving model used in this paper for Wimax MBS which is adapted from the model used for DVB-H in [1] and [5].

978-1-4244-2517-4/09/$20.00 2009 IEEE

Burst duration

applications like stock information broadcast may only require 32 to 64 kbps. From equation 5, the ratio of Rc/Rb is a critical component affecting the power savings hence Rb needs to be allocated sufficiently for a given Rc to achieve the targeted power savings percentage. Rb is dependent on the spectral efficiency (varies according to modulation/coding scheme, cell-size in SFN) and on the number of OFDMA symbols allocated for the MBS Zone. The other parameter affecting the power savings is the ratio (RcTsync/Bb) which incorporates the impact of Tsync. Bb is dependant on the number of OFDMA symbols allocated and the MCS used.
t MBS Zone

Burst Size

t Tb Toff Tsync Tb Ton MBS Burst Stream j MBS Burst stream j

Downlink subframe

Uplink subframe OFDMA Frame K + i

OFDMA Frame i

To compute Rb and Bb, we rely on the following information. The Modulation Coding scheme and spectral efficiency assumptions are listed in Table 1 while the system parameters are listed in Table 2. Note that for 1/4 guard interval, there would only be 40 OFDMA symbols available for data instead of 44 symbols for 1/8 guard interval.
Table 1: MCS and Spectral Efficiency
MCS QPSK (CTC) 1/2 16 QAM (CTC) 1/2 Number of Information bits in 1 PUSC Slot 48 96

Fig.1: Wimax MBS Power Saving Model

Ps =

T off

T on

T on + T off = T b + T sync

100

(1)

(2)

Rc
Tb
Bb
=

Bb T on + T off
Bb R b (1 h )

Table 2: SYSTEM PARAMETERS

Parameter Permutation Type BW FFT Size Cyclic Prefix OFDMA Frame Length Symbol Length Data Carriers Pilot Carriers Value PUSC 10 MHz 1024 1/8 5ms 102.56 720 120

(3)
(4)

Bb + Tsync Rb(1 h) Rc 100 = 1 Rc 1 + Tsync 100 Ps = Bb Rb(1 h) Bb Rc

(5)

Here Ps is the power savings in percentage, Ton is the time the receiver is turned on, Toff is the time the receiver is off, Tsync is the synchronization time required to transition from the Off to On state, Rc is the average stream bit rate of a MBS stream, Rb is the MBS Zone burst bitrate, h is the 802.16e protocol overheads for an MBS stream and Bb is the burst bit size. Note that Rc for a multimedia stream represents the sum of the composite audio and video streams plus the outer layer FEC redundancy information. Rb is also important for MBS radio resource allocation since it determines how many MBS streams can be supported. For example, in the case of N CBR (constant bitrate) encoded video streams with individual stream rates of RC,i the relationship with Rb is shown below:

The following equation captures the relationship between the spectral efficiency, number of OFDMA symbols and Wimax TDD frame duration to derive Rb.
Rb = Nbit N PUSC NMBS T frame (7)

Nbit is the number of information bits per PUSC slot, NPUSC is the number of PUSC slots per OFDMA symbol, NMBS is the number OFDMA symbols allocated for the MBS Zone per frame and Tframe is the duration of a Wimax TDD frame. We assume the subcarrier permutation used corresponds to PUSC all subcarriers and NMBS must be assigned in increments of two OFDMA symbols since one PUSC slot spans two OFDMA symbols per 802.16e standard. As an example, assuming PUSC all subcarriers with 10 MHz bandwidth, there are 30 PUSC slots in 2 OFDMA symbols for data. For 44 OFDMA symbols in one 5ms frame, MCS corresponding to QPSK (CTC) 1/2, the data rate corresponds to (48 x 15 x 44 / 0.005) 6.336 Mbps or 31680 bits (3960 bytes) per 44-symbol OFDMA frame. Based on the number of OFDMA symbols allocated for the MBS Zone, we can compute the burst size (Bb) and the MBS Zone data rate (Rb) that is used in the power savings computations.

R c , i = Rb(1 h)
i =1

(6)

The stream data rate Rc required to provide sufficient video quality is dependent on the characteristics of the source. For example, for sports channel on a QCIF or QVGA display, the average bitrate for H.264 encoded video should be a minimum of 256kbps, without other lower-layer protocol overheads and FEC overhead. For news and weather channel the video can be encoded at an average bitrate of 128 kbps, applications like audio streaming can be delivered using codecs like AAC with very good quality at 128 kbps and

100

P ow er S avi ngs P ercent age, P s ( ) %

1 T sync Ps = 1 Rc Rb (1 h ) + B b 100 Tsync 1 = 1 Rc + 1 R B (1 h ) ( D T ) R sync 1 1 2 R + (1 h ) R B c Tsync Rc D R 1 = 1 + 2 ( D R Tsync ) R B (1 h )

64 128 192 256 320 384 448 512

90 80 70 60 50 40

(9 )

100

M B S S t a m B i ra t , R c ( p s) re t e kb
B b 921 kbi , T b = 0. t 425 s, 16 sym bolZ one, Q P S K , 1/ G I 8 B b 1843 kbi , T b = 0. t 845 s, 16 sym bolZ one, Q P S K , 1/ G I 8 B b 460 kbi , T b = 0. t 215 s, 16 sym bolZ one, Q P S K , 1/ G I 8

The following figure provides an example of the power savings percentage versus the average receiving delay. Here we consider a 16 symbol/frame MBS Zone with MCS corresponding to QPSK CTC 1/2. Note that the burst duration is 0.635 seconds corresponding to a burst size of 1.448 Mbit.
P o w er S avin g s P ercen tag e, P s (% )

Fig.2: Power Savings Percentage versus Stream Data Rate (16 symbols/frame MBS Zone, QPSK CTC 1/2, Tsync 200ms, h 0.05)
Pow er Savings Per cent age,Ps (% )
100 95 90 85 80 75 70 65 64 128 192 256 320 384 448 512

100 90 80 70 60 50 40 0. 4 0. 8 1. 2 1. 6 2 2. 4 2. 8 3. 2 3. 6 4

A ve ra g e R e ce ivin g D e la y, D R (se c.)

R c 128 R c 256 R c 384 R c 512 kbps, R b kbps, R b kbps, R b kbps, R b 2. 3 2. 3 2. 3 2. 3 M M M M bps, 16 bps, 16 bps, 16 bps, 16 sym bolQ sym bolQ sym bolQ sym bolQ PSK PSK PSK PSK 1/ 1/ G 2, 8 1/ 1/ G 2, 8 1/ 1/ G 2, 8 1/ 1/ G 2, 8 I I I I

M B S S t a m B i ra t , R c ( p s) re t e kb
B b 1843 kbi , T b = 0. t 425 s, 16 sym bolZ one, 16 Q A M , 1/ G I 8 B b 3686 kbi , T b = 0. t 845 s, 16 sym bolZ one, 16 Q A M , 1/ G I 8 B b 921 kbi , T b = 0. t 215 s, 16 sym bolZ one, 16 Q A M , 1/ G I 8

Fig.4: Wimax MBS Power Savings Percentage versus Receiving Delay IV. CHANNEL SWITCHING DELAY TARGETS

Fig.3: Power Savings Percentage versus Stream Data Rate (16 symbols/frame MBS Zone, 16 QAM CTC 1/2, Tsync 200ms, h 0.05)

The figures above provide examples of the required resources. For stream data rates from 384 to 512 kbps using QPSK, even larger burst sizes (Bb) are not sufficient to provide power savings above 80%. For such scenarios, use of higher order modulations such as 16 QAM will be required but the ability to support 16 QAM will be limited to smaller cell-sizes due to the higher required C/I.
III. CHANNEL RECEIVING DELAY TARGETS

When the user switches channel from one MBS stream to another MBS stream, besides the delay to wait until the next burst on the target MBS stream, there are other additional delays that needs to be considered in order for the user to be able to view and hear the audio/video stream.

The next step is to describe the relationship between the channel switching delay and the power saving percentage. Here, we make use of the model from [5]. For the receiving delay, the average delay consists of two components namely the delay until the start of the burst and another delay until reception of the burst. The following equations describe the relationship:
1 DR = Ton + TOff + Ton = 2 =

Bb 1 Ton + Ton Ton + 2 Rc (8)

Fig.5: Mean Opinion Score versus channel change time study results from [7]

1 Bb 1 Bb Bb + Tb + Tsync = + + Tsync 2 Rc 2 Rc Rb (1 h)

Combining the results from equations 5 and 8, we obtain:

A useful reference to note is a presentation [7] made to ITU on IPTV Channel Zapping MOS evaluation. The figure above illustrates the relationship between MOS and Channel Zapping time for IPTV. We also note for comparison, performance targets from [4] indicate a channel switching

target of 2 seconds although we view this target should be less than 2 second per results from [7]. Note that we compare performance for Wimax MBS with IPTV since we assume RTP/UDP/IP as the protocol to transport the media streams. The receiving delay that was modeled in the previous sections accounts for variables like the MBS stream data rate, burst data rate and MBS Zone resource allocation size. However, there are other delays related to the video decoding process and audio/video synchronization that needs to be accounted for during channel switching. This section provides an overview and background of these delays. The I-frame rate impacts several key performance metrics namely the effective video compression ratio and the channel switching delay. For H.264 Level 1.2 video codec being considered for Wimax MBS, only I and P frames are used. The I-frame size is several times the size of P frames depending on content. Increasing the I-frame rates will reduce the compression efficiency while improving the channel switching delays. Note that the terms I-frame and IDR (Instantaneous Decoder Refresh) are used interchangeably in this paper. If we consider an example consisting of two group of pictures (GOP) namely I1P11P12P13P14P15P16P17P18P19P110 and I1P11P12P13P14P15I2P21P22P23P24P25, it is clear that the second GOP will consume a larger bandwidth while having a smaller switching time since the frequency of the I-frames is higher, hence on average a user switching to this channel will experience a shorter delay to receive the I-frame which also serve as a Random Access Point (RAP). The I-frame is required to decode the subsequent P-frames within a GOP, hence even if the user switches to another MBS channel it will need to wait for the I-frame first before it can begin decoding and viewing that MBS stream. From the video encoding part, the video encoder typically uses a bit budget to encode a video stream, hence if more IDR frames are used, then this will reduce the bit budget allocation to the P frames and subsequently reduce the video quality at the expense of a smaller channel switching delay. For Wimax MBS applications that are similar to IPTV i.e. there is an audio and video stream that are synchronized, the delay between initial tuning into the new channel and rendering the content of the new channel consists of these delay components [6]: Client stream buffering (network de-jitter buffer, video input buffer) Random access point acquisition Audio-video synchronization Mechanisms to minimize impact of packet-loss (FEC) Key acquisition times in case of scrambled content Next, we highlight the differences, optimizations and assumptions for the above delay components in Wimax MBS in comparison to wireline IPTV.

A. Client/MS Stream Buffering (DBuf): The terminal/MS needs to buffer incoming packets first, before forwarding to the video decoder to avoid under-run (due deviation in the arrival time of packets from the relative presentation timing), deliberately introduced jitter from the source, traffic smoothing/shaping and to compensate for network introduced jitter. For Wimax MBS, this delay is specifically for video streams and occurs during the time the client/MS fills its buffer before starting to render the video. For IPTV, this delay is in the range of 0.5 to 2 seconds from [6]. Here, we assume a minimum of 0.5 seconds delay is sufficient for mobile terminals. B. Random Access Point Acquisition (DRAP): The I-frame contains information that is necessary to decode the complete frame while the P-frames are differentially encoded and depends on the previously transmitted I-frames in order to decode. For reference, in 3GPP MBMS, the recommended range for the I-frame periodicity is from 500 ms to 4 seconds while current digital TV deployments over cable and satellite use a typical value of 600ms, corresponding to a GOP length of 15 frames at 25 fps [6]. We assume for Wimax MBS, the I-frame rate is 1000ms corresponding to a GOP length of 15 frames at 15 fps. In a corresponding wireline IPTV system, this would lead to an additional average RAP delay of 500ms.
One MBS Burst Random Access Point Delay

Channel B

Channel A

Time

Channel Switch Occurs Here

Fig.6: Channel switching between streams and Random Access Point Delay

However, in a Wimax MBS system, this depends on the MBS burst scheduling frequency for a particular stream. The figure above shows the equivalent IPTV RAP delay and also depicts an example whereby the I-frame is contained within one MBS burst transmission. From the example shown in Figure 4, the burst duration equals to 0.635 seconds and the total burst size spanning multiple OFDMA frames equals to 1448178 bits. If we assume 15 fps video stream, an I-frame average size of 2.2 kbyte and P-frame average size of 1.2 kbyte (assumption from [8]), this GOP average size would be 19456 bits. Since the total MBS burst size is a huge multiple of this number (e.g. 74 times average GOP size), one MBS burst will contain several GOP hence there will be more than one I-frame in the MBS burst. Hence, for Wimax MBS we dont need to account for an additional delay for the RAP delay in addition to the receiving delay as long as the following constraint ( N I

1 ) is met where NI is the

number of I-frames in a MBS burst. Based on the power savings requirement, this requirement is easily met since we need to schedule multiple GOP in a MBS burst to meet the power savings target e.g. Bb contains multiple GOP. C. Synchronization Between RTP Streams (DRTPSR): When Wimax MBS is used to send video streams, we assume RTP streaming is used. With RTP streaming, the audio and video is sent as separate RTP/UDP/IP streams. RTCP sender reports (SR) is required for synchronizing the audio and video streams and at least one RTCP SR has to be acquired for each stream to allow synchronized playouts of the streams. Some optimization is possible such as sending the RTCP SR when the I-frame is sent, which is our assumption here. Based on the similar constraint described previously ( N I 1 ) for DRAP, we do not need to account for additional delays for this component. D. Packet Loss Handling/FEC (DAL-FEC): We assume application FEC is used for the Wimax MBS (e.g. such as Raptor FEC). The FEC is applied to a block of packets to minimize the overhead and increase the interleaving depth. Packet loss is compensated by additionally transmitted FEC redundancy. To use the MBS data, decoding needs to be delayed until all packets of an application-layer FEC block are received. Here we assume an application layer FEC block of 0.5 Mbit is used. Based on the example in Figure 4 Bb is 1.44 Mbit which would not cause additional delay due to this component. Generally, as long as

Using a separate tune-in stream (with higher frequency of I-frames) in parallel to the main stream Using a low resolution tune-in stream and let device up-sample tune in stream Predictive tuning (client needs to receive and buffer predicted channels) Use of a channel change tune-in server located in the Access Network.
VII. CONCLUSIONS

Bb BALFEC (hence Tb DAL FEC ) where BAL-FEC is the application layer FEC block size, we do not need to account for this delay component in the overall switching delay.
E. Encryption Key Acquisition (DENC):: If the MBS content is encrypted, then the decryption keys must be acquired. There is an additional delay introduced by the key management if the key for the initial I-frame does not arrive during the client buffering and random access point acquisition procedures. Here, we assume the encryption key acquisition to occur in parallel and within the MBS burst duration. If DENC is greater than the MBS burst duration, then we need to account for this component delay. In summary, the total switching delay consists of many components and is summarized by equation 10 below:
DS = ( DR TB ) + max(TB , DRAP , DRTPSR , D AL FEC , DENC ) + DBuf (10)

In this paper, we identified key parameters impacting power management and channel switching issues for Wimax MBS services. Quantitatively, we evaluate the impact between the target power savings percentage, MBS Zone resource allocation and channel receiving delay. We also analyze and quantify the impact of various delay components besides the receiving delay that needs to be accounted for when switching video channels based on characteristics of video decoding procedures at the client/MS. A summary of optimization techniques from the wireline IPTV domain that can be applied to Wimax MBS were presented also. Optimization techniques to reduce the channel zapping delay are necessary in order to provide acceptable end-user experience for viewing video over Wimax MBS. Our analysis indicates the channel switching delay will be greater than 2 seconds if we target a power savings target of at least 80% with stream data rates between 256 kbps and 384 kbps with an MBS bearer rate of 2.3 Mbps using QPSK (CTC) 1/2 modulation coding scheme. The other option would be to use 16 QAM but this would be limited to smaller cell-sizes.
REFERENCES [1] ETSI, DVB-H Implementation Guidelines ETSI standard, TR 102.377, V1.2.1, 2005 [2] Air Interface for Fixed and Mobile Broadband Wireless Access Systems,IEEE Std 802.16e, February 2006 [3] HP iPAQ 210 Enterprise Handheld Review, Adama D. Brown, http://www.brighthand.com/default.asp?newsID=13810&review=HP+i PAQ+210+Enterprise+Handheld#Battery White Paper, FLO Technology Overview, [4] Qualcomm www.qualcomm.com [5] Rezaie, M., Bouaziz, I, Vadakital, V.K.M, Gabboui, M, Optimal Channel Changing Delay for Mobile TV Over DVB-H, IEEE International Conference on Portable Information Devices, Pg. 1-5, May 25 - 29 2007 [6] Fuchs, H., Farber, N, Optimizing channel change time in IPTV Applications, IEEE International Symposium on Broadband Multimedia and Broadcasting, 31 March 02 April 2008. [7] Admed Kamal, TNO, ITU-T IPTV Global Technical Workshop, Seoul, Korea, Oct. 13, 2006. [8] Wang, J., Venkatachalam, M. and Fang, Y., System Architecture and Cross-Layer Optimization of Video Broadcast over Wimax, IEEE Journal on Selected Areas of Communications, Vol. 25, No.4, May 2007.

Based on our previous assumptions and optimizations for the various delay components, equation 10 would become:
DS = DR + DBuf (11)

We note from the literature such as [6] optimization techniques that have been explored in the wireline IPTV domain that can also be adapted for Wimax MBS at the expense of more complexity and/or additional required bandwidth. These techniques include: