Chapter 1 Research Overview
1.3 Thesis Outline
In chapter 2, we will introduce the protocol of WIMAX MAC. In chapter 3, we will briefly introduce the scalable video coding (SVC). In Chapter 4, we will propose a mechanism and a cross-layer architecture to decide the proper bitrate of the coded video bitstream. In Chapter 5, we will simulate the performance of the proposed mechanism. In Chapter 6, we will present our conclusion and future work.
CHAPTER 2
OVERVIEW OF 802.16E SYSTEM
The mobile broadband wireless access system IEEE 802.16e [3], called IEEE 802.16e-2005 or mobile WIMAX, has been approved in December 2005. Mobile WIMAX is a enhancement to IEEE 802.16-2004 [4] to support subscriber stations moving at vehicular speeds. It specifies a system for combined fixed and mobile broadband wireless access, and tries to fill the gap between very high data rate WLAN and very high mobility cellular systems.
Because of the property of the high data rate, WIMAX system is expected to provide many kinds of services, like voice, internet, and multimedia services. For different services, they have different bandwidth request mechanism and different QoS definition according to the service characteristics. WIMAX system must do the scheduling well to meet the requirement of different services.
2.1 Introduction to PHY Layer of 802.16e PHY
There are three air interface defined in spec. One is WirelessMAN-SCa.
Another is WirelessMAN-OFDM. The other is WirelessMAN-OFDMA. And there are two duplex methods defined in spec. One is frequency division duplex (FDD).
The other is time division duplex (TDD). In our thesis, we adopt Wireless-OFDMA air interface and TDD mode. So the introduction to PHY Layer will be focused on the OFDMA PHY and TDD mode.
2.1.1 OFDMA PHY Frame Structure
FIGURE 1 WIMAX OFDMA FRAME STRUCTURE FOR TDD MODE
The OFDMA physical layer [3][4] can use the resource flexible and efficient because of the two dimension allocation. It supports TDD, Full and Half-Duplex FDD operation. TDD mode is much common for three reasons. Firstly, TDD mode can adjust the downlink/uplink ratio to support asymmetric downlink/uplink traffic load.
Secondly, TDD mode can provide global spectrum allocations. TDD mode just need a single channel for downlink and uplink. Downlink and uplink are separated from the guard time. The first part of transmission frame is downlink subframe, and the second part of transmission frame is uplink subframe. Thirdly, Transceiver designs for TDD mode implementations are less complex.
Now we will introduce the elements in the frame structure of Figure 1,
1) Preamble : The preamble occupies the first OFDMA symbol of the frame, and it is used for synchronization and channel estimation.
2) Frame Control Header (FCH) : The FCH contains DL_Frame Prefix. It specifies the length of the DL-MAP message and the repetition coding used for the DL-MAP message.
3) DL-MAP and UP-MAP : The DL-MAP and UL-MAP are used for resource allocation of DL and UL data bursts, and include burst-MS pairing information, the modulation and coding schemes of each data burst.
4) UL Ranging : The UL ranging channel is allocated for mobile stations to perform closed-loop time, frequency, power adjustment and bandwidth requests.
5) UL CQICH : The UL CQICH channel is allocated for mobile stations, mobile station can use it to feedback channel state information to base station.
6) UL ACK : The UL ACK channel is allocated for mobile stations, mobile stations use it to feedback DL HARQ acknowledge.
2.1.2 PHY Slot And Data Mapping
The OFDMA slot is a minimum unit for data transmission. One OFDMA slot occupies one subchannel and several OFDMA symbols. For downlink Full Usage of Subcarriers (FUSC) using the distributed subcarrier permutation, one slot is one subchannel by one OFDMA symbol. For downlink Partial Usage of Subcarriers (PUSC) using the distributed subcarrier permutation, one slot is one subchannel by two OFDMA symbols. For uplink Partial Usage of Subcarriers (PUSC) using the distributed subcarrier permutation, one slot is one subchannel by three OFDMA symbols. For downlink and uplink Band Adaptive modulation and coding (BandAMC) using the adjacent subcarrier permutation, one slot is one subchannel by one, two, three or six OFDMA symbols.
A Data Region is a two dimensional allocation which contents a group of contiguous subchannels and OFDMA symbols. All the allocation refers to logical subchannels. The minimum unit of data mapping is an OFDMA slot.
In Mobile WIMAX system, the BS decides that how many and which resource units are assigned to a connection. And the assignment of resource units is different in downlink and uplink. In downlink, system will consider the data size and try to fulfill the resource units in frequency domain first. After the frequency domain is fulfilled, and then system will try to fulfill the resource units in time domain. In uplink, the system will try to fulfill the resource unit in time domain first. After the time domain resource units are fulfilled, and then go to another frequency domain and repeat the procedure. Figure 2 and Figure 3 show downlink resource allocation and uplink resource allocation mechanism.
FIGURE 2 DL RESOURCE ALLOCATION
FIGURE 3 UL RESOURCE ALLOCATION
2.1.3 Subcarrier Permutation
Subcarrier permutation is a method to assign frequency subcarrier into subchannels. There are two kinds of permutation types. One is distributed subcarrier permutation. The other is adjacent subcarrier permutation.
Distributed permutation is that subcarriers belonging to a subchannel are selected pseudo randomly from all subcarriers. It can average intercell interference and avoid fading effect. The adjacent subcarrier permutation chooses adjacent subcarriers to form the subchannel. With adjacent subcarrier permutation, system can take advantage of frequency select fading and get multiuser diversity on the frequency domain.
IEEE 802.16e provides three ways to group subcarriers into the subchannel,
1) Full Usage of Subchannels (FUSC) :
This method is distributed permutation mode and is used in downlink only. It can use all subcarriers to do permutation for one subchannel and achieve the best frequency diversity by spreading subcarriers over entire band.
2) Partial Usage of Subchannels (PUSC) :
This method is distributed permutation mode and is used in both downlink and uplink. Firstly, subcarriers are grouped into several clusters. Then subcarriers are chosen one by one from each cluster to form a subchannel.
3) Band Adaptive Modulation and Coding (BandAMC) :
This method is adjacent permutation mode and is used in both downlink and uplink. The total bandwidth is divided into sub-bands and tries to utilize the frequency select fading to enhance the system performance.
2.1.4 Adaptive Modulation And Coding (AMC)
In Mobile WIMAX, the system will adaptively change the modulation coding scheme according to the channel condition of the radio link. The system supports severval modulation types : Quadrature Phase Shift Keying (QPSK), 16-state Quadrature Amplitude modulation (16-QAM), and 64-state Quadrature Amplitude modulation (64-QAM). And the system also supports several coding schemes : Convolution Code (CC), Low Density Parity Check Code (LDPC), Block Turbo Code (BTC), and Convolution Turbo Code (CTC). TABLE 1 summarizes the modulation coding scheme supported in the Mobile WIMAX profile.
TABLE 1 SUPPORTED CODE AND MODULATION
DL UL
Modulation QPSK,16QAM,64QAM QPSK,16QAM,64QAM
Code
Rate
CC 1/2, 2/3, 3/4, 5/6 1/2, 2/3, 5/6 CTC 1/2, 2/3, 3/4, 5/6 1/2, 2/3, 5/6 Repetition x2, x4, x6 x2, x4, x6 2.2 Introduction to MAC Layer of 802.16e MAC 2.2.1 Layer Structure
FIGURE 4 PHY-MAC STRUCTURE IN IEEE 802.16E
As shown in Figure 4, there are three sublayers in the MAC layer. We will introduce the functionality of each sublayer.
1) Service-Specific Convergence Sublayer (CS) : This sublayer is an interface between upper layer and MAC layer. The most important functionality of CS is to identify different traffic from upper layer and to assign connection ID (CID) to each connection.
2) Common Part Sublayer (CPS) : This sublayer manages the main function of controlling the whole radio resource, such as QoS control, fragmentation, packing, scheduling, request-and-grant, admission control, handover and QRQ.
3) Security Sublayer : This sublayer performs the authentication of network access, registration, key exchange and encryption of PDUs.
2.2.2 MAC PDU Formats
The MAC PDU is a data unit between the BS MAC layer and MS MAC layer.
A MAC PDU consists of a 48 bits MAC header, a variable length data payload, and an optional 32 bits Cyclic Redundancy Check (CRC). Some MAC PDU will not include payload and CRC bits. These kinds of PDUs are used only in the uplink to transmit control message. These MAC signaling headers include bandwidth request, uplink transmit power report, CINR report, CQICH allocation request, PHY channel report, uplink sleep control, SN report, and feedback functionalities. MAC PDUs also include some subheaders. Those subheaders will be inserted in MAC PDUs following the generic MAC header. Those subheaders help system perform grant management, packing, ARQ feedback, and so on.
2.2.3 Fragmentation And Packing
In WIMAX system, the MAC SDU coming from CS will be formatted according to the MAC PDU format in the CPS, possibly with fragmentation and packing due to efficient utilization of the radio resource and packet error rate.
Fragmentation process is to divide a SDU into several PDUs payload areas.
There are two reasons for fragmentation. One is that the SDU size is larger than the maximum size of PDU payload. The other is for preventing high packet error rate.
The larger PDU size is, the higher packet error rate is. So the WIMAX system needs to divide the SDU size properly according to the channel condition. Figure 5 shows the process of fragmentation.
Packing process is to pack several SDUs into a single PDU payload. In this way, system may avoid resource waste due to the overhead caused by MAC header and CRC. Figure 6 shows the process of packing.
FIGURE 5 FRAGMENTATION
FIGURE 6 PACKING
2.2.4 QoS Based Service Classes
The IEEE 802.16e standard provides several QoS classes for different kinds of services. For different QoS classes, system sets different parameters and transmission/request methods to meet the requirement of different kinds of service.
Here will introduce these classes:
1) Unsolicited Grant Service (UGS) : Designed to support real-time service flows that generate fixed-size data packets periodically, such as T1/E1 and VoIP without silence suppression.
2) Real-time Polling Service (rtPS) : Designed to support real-time service flows that generate variable-size data packets, such as Moving Picture Experts Group (MPEG) video.
3) Extended Real-time Polling Service (ertPS) : A scheduling mechanism that builds on the efficiency of both UGS and rtPS. The BS provides unsolicited unicast
grants as in UGS, thus saving the latency of a bandwidth request. However, UGS allocations are fixed in size, whereas ertPS allocations are dynamic.
4) Non-real-time Polling Service (nrtPS) : The nrtPS is designed for non-real-time service that can tolerate more delay, such as FTP, web-browsing and so on.
5) Best Effort Service (BE) : BE service is with the lowest QoS level. These kinds of service are designed to support data streams for which no minimum service level is required and therefore may be handled on a space-available basis.
CHAPTER 3
OVERVIEW OF SCALABLE VIDEO CODING
3.1 Introduction to Scalable Video Coding
Data networks for video communication are growing fast nowadays. The environment varies from broadband cable/ADSL networks to wireless/mobile networks. Besides, the display monitors of the devices are also diversified. It may be a small size screen on a mobile device or a high definition projection system. For different applications on various devices or under different network conditions, the available bandwidth and resource may be highly divergent. To overcome different application scenarios, the idea of scalable video coding is proposed.
The scalable video coding (SVC) standard [5] is an extension of the H.264/AVC standard [6] developed by the Joint Video Team (JVT) that uses a single bit-stream to provide multiple frame rates, frame sizes and quality levels while achieving a reasonable coding efficiency.
3.2 Encoder Overview
The SVC encodes the video into multiple spatial, temporal, and SNR layers for combined scalability. Figure 7 shows the generic structure of an SVC encoder with three spatial layers.
Figure 7 SVC encoder structure with three spatial layers [7]
SVC encoder provides three different scalable features in spatial, temporal and SNR layer respectively. Spatial scalability and CGS are achieved by multiple layers with a pyramid structure. Temporal scalability is achieved by a temporal decomposition using hierarchical B pictures. FGS is achieved by encoding successive refinements of the transform coefficients.
3.3 Hierarchical-B Prediction Structure
SVC encoder uses hierarchical-B prediction structure to achieve multilevel temporal scalability. Figure 8 depicts a hierarchical-B prediction structure with 4 temporal levels and a GOP size of 8. Each key picture is either an intra-coded frame(I frame) or a P frame that uses the previous key picture as the reference picture. The picture number from 1 to 7 are B frames. Each B-frame is bi-directionally predicted
using both previously and future displayed reference pictures from the lower temporal level. The hierarchical-B structure has better coding efficiency using more efficient frame level bit allocation, especially for sequences with fine texture and regular motion.
Figure 8 Hierarchical-B prediction structure [7]
3.4 Inter-layer Prediction Structure
Figure 9 Inter-layer prediction structure with three spatial layers [7]
Interlayer prediction is dependent on the types of layers used. The spatial and CGS layers can flexibly select the reference layer from any lower layers while the FGS layer must be predicted from the previous SNR layer at the same resolution. As demonstrated by an example in Figure 9, the three columns represent three spatial resolutions: QCIF, CIF, and 4CIF. Each spatial resolution contains several SNR layers, and the arrow specifies the reference layer.
CHAPTER 4
THE PROPOSED CROSS-LAYER DESIGN
In this chapter, we will propose a mechanism and a cross-layer architecture between the SVC system and the WIMAX MAC to solve the problem discussed in chapter 1.1 and chapter 1.2.
4.1 The Design Issues
Bandwidth fluctuation in Mobile WIMAX system is caused by the variation of bandwidth allocation and the channel condition. And it brings many problems during the transportation of real time video bitstream (video conference or video telephone), such as video latency, available bandwidth wastage and buffer condition unstable. So the first design issue is latency issue, and the second design issue is the utilization of the available bandwidth.
What we have to do is to estimate a proper bitrate of the coded video bitstream for each GOP and send it to the SVC extractor. With this bitrate, the SVC extractor will extract the proper size bitstream for each GOP to transmit. So during the transportation of real time video bitstream, we could have the better performance in video latency in the receiver, and also have the efficient utilization of available bandwidth in the transmitter.
4.2 The Proposed Cross-Layer Architecture
FIGURE 10 THE PROPOSED CROSS-LAYER ARCHITECTURE
The most important component of the proposed architecture is the Bitrate Decision Engine. How to design this component will be discussed in chapter 4.3. Now we should focus on the overall coded video bitstream flow and control information flow of the proposed cross-layer architecture, as shown in Figure 10.
The video source is sent to SVC encoder, and then the coded bitstream is sent to the SVC extractor. Before the beginning of each GOP, SVC extractor will ask Bitrate Decision Engine to feedback the extracted video bitrate, and uses this bitrate to extract the proper size of bitstream for this GOP.
There is one buffer between MAC and SVC extractor. It is FIFO buffer. When the SVC extractor starts to extract the SVC bitstream, it will ask MAC to transmit the data in buffer at the same time.
There are two inputs of the Bitrate Decision Engine. One is the size of data stored in buffer. The other is the available bandwidth information of the period in the past. With the input information, the bitrate decision engine will decide the bitrate of the coded video bitstream for the present GOP.
4.3 The Bitrate Decision Engine
In chapter 4.2, we have mentioned that SVC extractor will ask Bitrate Decision Engine to feedback the extracted video bitratebefore the beginning of each GOP. To achieve a better performance, an enhanced feedback control mechanism will be proposed.
Before the discussion of the proposed enhanced feedback control mechanism, we will first discuss four different conventional feedback control mechanisms.
Four different conventional feedback control mechanisms:
1) Mean Mechanism: There are 53 transmission frames (one frame is 5 ms in WIMAX system) in each GOP period (one GOP period is 8/30 seconds in our thesis). The Mean Mechanism uses the average transmission bitrate of the last GOP period to be the extracted video bitrate for the present GOP period. The formula is shown in eq(1).
EN = DN + ⋯ + DN−52
T × 53 (1)
2) Median Mechanism: There are 53 transmission bitrates for each transmission frame in the last GOP period. And the Median Mechanism uses the median of these transmission bitrate to be the extracted video bitrate for the present GOP period. The formula is shown in eq(2).
EN = median of DN−52
T 、… 、DN
T (2)
3) IIR Mechanism : IIR Mechanism is adopted by [1]. The formula is shown in eq(3).
EN = p × EN−1+ (1 − p) × DN
T (3)
4) Instant Mechanism: Instant Mechanism uses the first transmission bitrate in the present GOP period to be the extracted video bitrate for the present GOP period. The formula is shown in eq(4).
EN =DN
T (4)
where
EN: the extracted video bitrate for the present GOP period (bits/sec) EN−1: the extracted video bitrate for last GOP period (bits/sec)
DN : the data size which can be transmitted at the first transmission frame of this GOP period (bits)
DN−i: the data size which can be transmitted at the 53 − i th transmission frame
of the last GOP period (bits)
p: 0.2
T: the time of one transmission frame (sec)
Enhanced feedback control mechanism(proposed):
Enhanced feedback control mechanism is proposed to satisfy the latency issue. It modifies the extracted video bitrate decided by conventional feedback control mechanism based on the data size stored in the buffer. The modification size is shown in eq(5) :
∆ = α × DB
N × T (5)
where
∆∶ the modification size (bits/sec)
DB: the data stored in the buffer which the MAC is transmitting (bits)
T: the time of one transmission frame (sec)
N: the number of transmission frames in one GOP period
α: the cofficient of the modification size, 0 ≤ α ≤ 1. In our thesis, α is set to 0.3 by experiment
Finally, the Bitrate Decision Engine will send the modified extracted video bitrate to the SVC extracter. The formula is shown in eq(6) :
Modified extracted video bitrate = EN − ∆ (6)
where
EN: the extracted video bitrate decided by conventional feedback control mechanism
for the present GOP period (bits/sec)
4.4 Discussion: The reason why enhanced feedback control mechanism can improve the performance of latency issue
Sometimes, the extracted SVC data which can’t be transmitted to the receiver in its GOP period will be stored in MAC buffer, and it will be transmitted during the next GOP period. The data which can’t be transmitted to the receiver in its GOP period is named Delay Data.
There are two types of data which need to be transmitted during the present GOP period. One is the Delay Data. The other is the extracted SVC bitstream of the present GOP. Therefore, we need to divide the transmission bitrate of the present GOP period into two parts. One part is used to transmit the extracted SVC bitstream of the present GOP. The other part is used to transmit the Delay Data.
The transmission bitrate for the present GOP period, EN , is decided by conventional feedback control mechanism, and we divide it into two parts, ∆ and EN − ∆. The ∆ decided by enhanced feedback control mechanism is used to transmit the Delay Data. EN− ∆ is used to transmit the extracted SVC bitstream of the present GOP.
So the reason why enhanced feedback control mechanism can improve the performance of latency issue is because we reserve one part of the transmission bitrate,
∆ , and use it to transmit the Delay Data.
CHAPTER 5
SIMULATION SETUP AND RESULTS
5.1 Simulation Setup 5.1.1 Cell Plane
FIGURE 11 CELL PLANE
As shown in Figure 11, there are 19 cells in the simulation platform. The cell radius is 1km. And the resue factor and BS sector are both three. The interference cells are marked red color. So there are six interference links and one useful link
As shown in Figure 11, there are 19 cells in the simulation platform. The cell radius is 1km. And the resue factor and BS sector are both three. The interference cells are marked red color. So there are six interference links and one useful link