Evaluation of the strategies for error resilient transmission of 3D data through a transmitter-channel-receiver pipeline Gozde B Akar M.

Transcription

1 Evaluation of the strategies for error resilient transmission of 3D data through a transmitter-channel-receiver pipeline Gozde B Akar M. Oguz Bici Anil Aksay Done Bugdayci Antti Tikanmäki Atanas Gotchev

2 Project No Evaluation of the strategies for error resilient transmission of 3D data through a transmitter channel receiver pipeline Gozde B Akar, M. Oguz Bici, Anil Aksay, Done Bugdayci, Antti Tikanmäki, Atanas Gotchev Abstract: In this report, we study error resilient approaches aimed at better error protection of stereo video content to be transmitted over DVB H. We focus on channel models described in the recent DVB H implementation guide. For this models, we run extensive simulations to quantify the advantages and disadvantages of different protection schemes for different video bit rates. Our error protection schemes are based on the standard for DVB H. DVB H uses FEC for error protection at physical layer and comes with an optional FEC tool (MPE FEC) at the link layer. In our approach, we propose to use the a priori knowledge of the transmitted media and apply MPE FEC intelligently to provide better robustness. That is, we investigate Unequal Error Protection (UEP) favouring the more important content. Extensive simulation results are given to show the effect of MPE FEC and UEP on 3D coded video under different channel models, conditions and bitrates. Keywords: Error characteristics, stereo, error resilience, UEP, EEP

3 Executive Summary In this report, we aim at quantifying the effect of the link-layer MPE-FEC tool on the transmission of stereo video over DVB-H. The FEC uses Reed-Solomon (RS) FEC codes encapsulated into Multiprotocol encapsulated sections (MPE-FEC). The MPE-FEC is supposed to improve the carrier-to noise (C/N) and Doppler performance in the DVB-H channel while also providing improved tolerance of impulse interference. However, MPE-FEC might fail in the presence of very erroneous conditions. In our study, we consider typical mobile channel models and stereo video bit rates at different channel conditions by varying SNR and other channel parameters. We develop 3D video specific error resilient techniques where we utilize a priori knowledge of the transmitted media. Namely, we apply MPE-FEC intelligently to provide better robustness by assigning different protection rates to differently-important content, the so-called Unequal Error Protection (UEP) mechanism. We present extensive simulation results which demonstrate the feasibility of our approach. Page 2 of 44

4 Table of Contents Executive Summary Introduction Error-Resilient Transmission of Conventional Video over DVB-H New models of the DVB-H transmission channel Physical Layer (TS) Statistics Application Layer Statistics Operational Modes Existing Error Protection Approaches for Conventional Video over DVB-H Error resilient transmission of 3D video data over DVB-H End-to-end simulation test system UEP Approaches for Stereoscopic Video Transmission over DVB-H Experimental Results Conclusions Acknowledgements References Page 3 of 44

5 Abbreviations ADT AL FEC BER C/N DVB DVB-H DVB-T FEC FFT IP IPDC MFER MPE MPE FEC MPE ifec MPEG PID QAM QPSK RS RSDT TS WING-TV PI PO Application Data Table Application Layer Forward Error Correction Bit Error Ratio Carrier to Noise ratio Digital Video Broadcasting DVB Handheld DVB Terrestrial Forward Error Correcting code Fast Fourier Transform Internet Protocol IP Datacasting MPE-FEC Frame Error Ratio Multi Protocol Encapsulation Multi Protocol Encapsulation - Forward Error Correction Multi Protocol Encapsulation inter Burst Forward Error Correction Moving Picture Experts Group Packet IDentifier Quadrature Amplitude Modulation Quadrature Phase Shift Keying Reed-Solomon Reed-Solomon Data Table Transport Stream Services to Wireless, Integrated, Nomadic, GPRS-UMTS & TV Handheld Terminals Pedestrian/Portable Indoor Pedestrian/Portable Outdoor Page 4 of 44

6 1 Introduction Wireless networks are often error prone due to factors such as multipath fading and interferences. In addition, the channel conditions of these networks are often non-stationary, such that the available bandwidth and channel error rates are changing over time with large variations. In order to maintain satisfactory QoS, a number of technologies have been proposed targeting different layers of the networks. Concentrating more on DVB-H, it uses FEC for error protection and comes with an optional FEC tool at the link-layer. This FEC uses Reed-Solomon (RS) FEC codes encapsulated into Multi-protocol encapsulated sections (MPE-FEC). The MPE-FEC was also introduced to provide additional robustness required for hand-held mobile terminals. MPE-FEC improves the carrier-to noise (C/N) and Doppler performance in the DVB-H channel while also providing improved tolerance of impulse interference. In this report, we study the impact of MPE-FEC over the quality of the delivered stereo video for different channel conditions. The models used for the channel are obtained from the DVB-H Implementation Guideline [4] and Wing-TV [3]. These models are named according to the environment: Indoor Commercial and Outdoor Residential by JTC, and Portable Indoor (PI), Portable Outdoor (PO), Vehicular Urban and Motorway by Wing-TV. As shown in the experimental results, although MPE-FEC provides the much needed data robustness for 3D video transmission in wireless channels, under very erroneous conditions it may fail. In order to overcome this we use a-priori knowledge of the media to differentially protect data using FEC. The results show a clear improvement on the received video quality when UEP is used. This report is arranged as follows: Section 2 gives the new models of the DVB-H transmission channel together with comparisons of the reported Wing-TV results and the results of our simulator in terms of ABEL, PER, VBEL, etc; the available operational modes and error protection schemes for DVB-H. Section 3 gives our approach for error resilient 3D transmission of video over DVB-H together with the proposed UEP strategies and the experimental results. 2 Error-Resilient Transmission of Conventional Video over DVB-H 2.1 New models of the DVB-H transmission channel There are several channel models developed by different communities and researchers for representing RF propagation characteristics namely; Typical Urban (6 and 12 taps), Bad Urban (6 and 12 taps), Rural Area (4 and 6 taps), Hilly Terrain (6 and 12 taps) by COST 207 [1]; Indoor Commercial (7 taps), Outdoor Residential (10 taps) by JTC [2] and Pedestrian Indoor, Pedestrian Outdoor, Vehicular Urban, Motorway Rural by Wing-TV [3]. Among these models pedestrian indoor, pedestrian outdoor and typical urban 6 taps are included in the latest version of DVB BlueBook for DVB-H Implementation Guide as Portable Indoor, Portable Outdoor and Mobile Channel [4]. Expected DVB-H receiver performance when noise (N) is applied together with the wanted carrier (C) using a degradation point criteria MFER 5 % is given in Table 2-1. Page 5 of 44

7 Modulation Convolutional Code Rate MPE-FEC Rate QPSK 1/2 1/2 6,6 7,6 QPSK 1/2 2/3 6,8 7,8 QPSK 1/2 3/4 7 8 QPSK 1/2 5/6 7,2 8,2 QPSK 1/2 7/8 7,4 8,4 QPSK 2/3 2/3 9,8 10,8 QPSK 2/3 3/ QPSK 2/3 5/6 10,2 11,2 QPSK 2/3 7/8 10,4 11,4 16-QAM 1/2 2/3 12,8 13,8 16-QAM 1/2 3/ QAM 1/2 5/6 13,2 14,2 16-QAM 1/2 7/8 13,4 14,4 16-QAM 2/3 2/3 15,8 16,8 16-QAM 2/3 3/ QAM 2/3 5/6 16,2 17,2 16-QAM 2/3 7/8 16,4 17,4 PI PO Table 2-1 C/N (db) for 5 % MFER in PI & PO channel We have implemented these models in the Simulink model using the Multipath Rayleigh Fading Channel (MRFC) block with the tapped-delay-line delay and gain values are set as specified in the literature [4]. The Doppler spectrum of each tap is also set as specified in standard. For the MR and VU models the Doppler spectra are of either the Gaussian or the classical Jake s type. The pedestrian models (PI and PO), on the other hand, have sum of Gaussian and delta function as the Doppler spectrum of their first tap meaning that there exists a line-of-sight component between transmitter and receiver. This behavior is approximated by a BiGaussian distribution where one of the Gaussians approximates the Dirac delta function by having very small standard deviation. This approximate behavior of PI & PO channel models of Simulink simulator delivers different error patterns when compared to error trances generated e.g. by Wing-TV simulators. Both simulations, however, work with simplified receiver models. The block diagram of the DVB-T physical layer simulator can be seen in Figure 2-1. In practice, in receiver side, after guard interval removal and OFDM demodulation, receiver calculates an estimate of the channel by looking at the pilot carriers. Pilot carriers have modulation parameters known to the receiver, so according to the changes on pilot carriers, the receiver runs a channel estimation parameter. There are several channel estimation techniques for OFDM systems and a lot of research is done related to the problem. Since DVB standards do not specify the receiver structure, performance of the receiver due to channel estimation is highly implementation dependent. In the available Simulink model, the Pilot Processing Block right after the OFDM receiver block assumes perfect channel knowledge. Channel estimation can be included in the simulation according to the specifications of the receiver architecture that will be used in hardware device. The Wing TV simulations have been implemented with channel estimation by linear interpolation implemented in two steps, one in time domain and one in frequency domain. It should be noted that this is still a simpler model as more complicated channel estimation techniques are implemented in receiver architectures in hardware. Page 6 of 44

8 Figure 2-1 Simulink model for DVB-H physical layer simulation Physical Layer (TS) Statistics In this section we provide a comparison of the simulators using WING-TV and Simulink model traces. Wing-TV traces are TS packets long and the Simulink traces varied in length between and TS packets. Physical layer parameters of the traces are given in Table 2-2. Modulation 16-QAM FFT size 8K Guard interval 1/4 Convolutiona Code Rate 2/3 Bandwidth 8 MHz Table 2-2 Physical layer parameters of the traces used to extract statistics Page 7 of 44

9 TS-PER TS-PER MOBILE3DTV Packet Error Rate (PER) vs. channel SNR Motorway 100Hz - 16QAM 8K 2/3 (unprocessed error trace) wingtv simulink SNR Figure 2-2 Motorway Rural Model TS-PER comparison Results Vehicular Urban 30Hz - 16QAM 8K 2/3 (unprocessed error trace) wingtv simulink SNR Figure 2-3 Vehicular Model TS-PER comparison Results Page 8 of 44

10 TS-AEBL TS-PER MOBILE3DTV Pedestrian Indoor 10Hz - 16QAM 8K 2/3 (unprocessed error trace) wingtv simulink SNR Figure 2-4 Pedestrian Indoor Model TS-PER comparison results Average Error Burst length (ABEL) vs. Packet error rate (PER) Motorway 100Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-5 Motorway Rural Model TS-ABEL comparison Results Page 9 of 44

11 TS-AEBL TS-AEBL MOBILE3DTV Vehicular Urban 30Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-6 Vehicular Urban Model TS-ABEL comparison Results 25 Pedestrian Indoor 10Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-7 Pedestrian Indoor Model TS-ABEL comparison Results Page 10 of 44

12 TS-VEBL TS-VEBL MOBILE3DTV Variance of Error Burst Length (VBEL) vs. Packet Error Rate (PER) Motorway 100Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-8 Motorway Rural Model TS-VBEL comparison Results Vehicular Urban 30Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-9 Vehicular Urban Model TS-VBEL comparison Results Page 11 of 44

13 TS-ATBE TS-VEBL MOBILE3DTV 4.5 x Pedestrian Indoor 10Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-10 Pedestrian Indoor Model TS-VBEL comparison Results Average Time between Errors (ATBE) vs. PER 10 4 Motorway 100Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-11 Motorway Rural Model TS-ATBE comparison Results Page 12 of 44

14 TS-ATBE TS-ATBE MOBILE3DTV 10 4 Vehicular Urban 30Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-12 Vehicular Urban Model TS-ATBE comparison Results 10 4 Pedestrian Indoor 10Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-13 Pedestrian Indoor Model TS-ATBE comparison Results Page 13 of 44

15 TS-VTBE TS-VTBE MOBILE3DTV Variance of Time between Errors (VTBE) vs. PER 10 8 Motorway 100Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-14 Motorway Rural Model TS-VTBE comparison Results 10 8 Vehicular Urban 30Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-15 Vehicular Urban Model TS-VTBE comparison Results Page 14 of 44

16 TS-VTBE MOBILE3DTV Pedestrian Indoor 10Hz - 16QAM 8K 2/3 (unprocessed error trace) simulink wingtv TS-PER Figure 2-16 Pedestrian Indoor Model TS-VTBE comparison Results Application Layer Statistics The error bursts are measured using number of contiguously lost NALUs, which in this case is equal to the number of video frames. The simulation was performed by dropping transport stream packets if they were marked as lost in the error trace. The transport stream used in the simulations contained a stereoscopic MVC video stream. Each video frame was coded as one slice (and therefore in one NALU) and split into several RTP packets if its size exceeded the MTU of 1500 bytes. In the experiment, the IP packets in the received TS are decapsulated, the RTP packetization is decoded, and finally information of lost and received NALUs is collected. Page 15 of 44

17 Average Error Burst Length (video frames) vs. TS-PER Figure 2-17 Motorway Rural Model NALU-ABEL comparison Results Figure 2-18 Vehicular Urban Model NALU-ABEL comparison Results Page 16 of 44

18 Average Time between Errors (video frames) vs. TS-PER Figure 2-19 Motorway Rural Model NALU-ATBE comparison Results Figure 2-20 Vehicular Urban Model NALU-ATBE comparison Results Page 17 of 44

19 Standard Deviation of Error Burst Lengths (video frames) vs. TS-PER Figure 2-21 Motorway Rural Model NALU-stdBEL comparison Results Figure 2-22 Vehicular Urban Model NALU-stdBEL comparison Results Page 18 of 44

20 Standard Deviation of Time between Errors (video frames) vs. TS-PER Figure Motorway Rural Model NALU-stdTBE comparison Results Figure 2-25 Vehicular Urban Model NALU-stdTBE comparison Results Page 19 of 44

21 2.2 Operational Modes DVB-T physical layer structure provides numerous combinations for the transmission scheme where parameters like modulation type, convolutional code rate, guard interval, bandwidth, FFT size, and interleaving mode can be set to different options. In addition to these physical layer parameters there are a number of combinations introduced at the link layer too. Link layer parameters are basically the burst parameters and delta-t. Figure 2-26 Burst Parameters In Figure 2-1 the relationship between burst parameters are illustrated. Here it should be noted that not all of the burts parameters are independent. Once the physical layer parameters modulation type, convolutional code rate and guard interval are chosen, the burst bit rate is set to a constant value according to Table 2-2 [8]. Then for a fixed row size chosen among four possibilities (256, 512, 768, 1024) the burst duration is figured. In DVB-H typical implementations, strong code rates like 1/2, 2/3 and longest guard intervals as 1/4, 1/8 is utilized for a robust transmission scheme. Modulation Convolutional Code Rate QPSK 16-QAM Guard Interval 1/4 1/8 1/16 1/32 1/2 4,98 5,53 5,85 6,03 2/3 6,64 7,37 7,81 8,04 3/4 7,46 8,29 8,78 9,05 5/6 8,29 9,22 9,76 10,05 7/8 8,71 9,68 10,25 10,56 1/2 9,95 11,06 11,71 12,06 2/3 13,27 14,75 15,61 16,09 3/4 14,93 16,59 17,56 18,1 5/6 16,59 18,43 19,52 20,11 7/8 17,42 19,35 20,49 21,11 Table 2-3 Useful bit rate (Mbit/s) for all combinations of guard interval, constellation and code rate for non-hierarchical systems for 8 MHz channels (irrespective of the transmission modes) Page 20 of 44

22 2.3 Existing Error Protection Approaches for Conventional Video over DVB-H H.264/AVC RTP UDP IP MPE MPEG2-TS DVB-T Table 2-4 Layered structure of IPDC over DVB-H for mobile TV DVB-H standard carries the user data in IP datagram (IP Data Casting IPDC) which are encapsulated into MPE (Multi Protocol Encapsulation) sections and mapped to MPEG-2 transport streams to be used with DVB-T physical layer. DVB-T physical layer provides error protection through inner (RS) and outer (punctured convolutional) coding and inner (native or in-depth) and outer (convolutional) interleaving mechanisms. In addition to physical layer error protection, in order to improve the C/N and Doppler performance in mobile channels and to increase tolerance to impulse interference, an optional forward error protection mechanism called MPE-FEC is included at the link layer. When MPE-FEC is employed, the FEC frame consists of ADT and RSDT having a constant number of columns equal to 255. The number of rows of a FEC frame can be either 256,512,768 or RS codewords are obtained from ADT in a row-wise fashion and RSDT is filled by the parity bytes accordingly. Each IP datagram is transmitted in an MPE section and each column of the RSDT is transmitted in an MPE-FEC section. MPE-FEC provides error correction through RS coding and time interleaving of IP datagram and sending the data and related parity in one burst. In other words MPE and MPE-FEC sections of a FEC frame are transmitted as MPE sections followed by MPE-FEC sections. So in a lossy channel transmission experience if a burst is partially received and partially lost; the lost parts may be recovered according to the percentage of the loss. However recovery from a completely lost burst is not possible. To make recovery from multiple successive complete burst losses possible, one solution is to encode information from several consecutive bursts in a jointly manner. The easiest way of implementation is to consider the user data of several bursts as one large data block and compute parity bytes from this large data block. Later the parity data can be either distributed into several bursts or sent after the data block. Different strategies for the distribution of parity bytes over bursts are studied and performance evaluation is provided in [9]. DVB-H standard adopts multi burst protection for the file delivery services by means of replacing the link layer MPE-FEC mechanism with another FEC mechanism at the application layer using systematic Raptor coding, which is referred as AL-FEC. Multi-burst protection outperforms singleburst protection in terms of coding efficiency with the increased time interleaving of the encoded data and increased spatial diversity of the code. However the increased time interleaving introduces increased network latency, which is the main drawback of this approach. It has been shown that for multimedia streaming services the FEC efficiency can be considerably improved by delivering the streaming content as a succession of larger data blocks [7]. Furthermore the robustness of the DVB-H transmission increases not only as FEC overhead is increased but also as the number of bursts jointly encoded increased. But of course, as the number of jointly encoded Page 21 of 44

23 bursts increase the introduced network latency becomes the bottleneck, which in the case of mobile TV is the maximum zapping time between channels that can be tolerated. However it should be kept in mind that for channel conditions enabling good signal reception information bytes are distributed similarly across the bursts for both AL-FEC and MPE-FEC utilized, meaning that the receiver can play the contents of a burst as soon as it receives that burst without waiting for succeeding jointly coded bursts. The latency mentioned above is a matter of discussion when FEC decoding is necessary due to relatively worse reception conditions. Hence in order to compare the two protection mechanisms, conventional streaming delivery approach with MPE-FEC and multiburst protection approach with AL-FEC, results should be evaluated at the erroneous channel conditions and as a function of the coding rate and the network latency introduced. For a detailed analysis of comparison of these two strategies user can refer to article provided in [7]. Alternative to AL-FEC, it is possible to implement Raptor coding at the link layer using the MPEiFEC (MPE inter-burst FEC) protection mechanism of DVB-SH [6] which is defined backwards compatible with DVB-H. It is also possible to use the same RS code adopted in MPE-FEC but with a sliding window coding approach, called Sliding Reed-Solomon Encoding (SRSE). 3 Error resilient transmission of 3D video data over DVB-H 3.1 End-to-end simulation test system The building blocks of the system used in this report can be seen Figure 3-1. Figure 3-1 The overall system Stereo video content with right and left view is first compressed with a stereo video encoder (joint or simulcast). Resulting Network Abstraction Layer (NAL) units (NALU) are fed to the stereo video streamer. The streamer encapsulates the NAL units into Real Time Transport Protocol (RTP), User Datagram Protocol (UDP) and finally Internet Protocol (IP) datagram for each view separately. The resulting IP datagram are encapsulated to MPE sections each of which consisting of a header, the IP datagram as a payload, and a 32-bit cyclic redundancy check (CRC) for the verification of payload integrity. On the level of the MPE, an additional stage of forward error correction (FEC) can also be added. This technique is called MPE-FEC and improves the C/N and Doppler performance in mobile channels. To compute MPE-FEC, IP packets are filled into an N x 191 matrix where each square of the matrix has one byte of information and N denotes the number of rows in the matrix. The standard defines the value of N to be one of 256, 512, 768 or The datagrams are filled into the matrix column-wise. Error correction codes (RS codes) are computed for each row and concatenated such that the final size of the matrix is of size Nx255. To adjust the Page 22 of 44

24 effective MPE-FEC code rate, padding or puncturing can be used. Padding refers to filling the application data table partially with the data and the rest with zero whereas puncturing refers to discarding some of the rightmost columns of the RS-data table. The IP input streams coming from different sources are encapsulated and multiplexed for transmission according to the time slicing method. Figure 2 illustrates the MPE-FEC structure. The left and right views are assigned different PIDs and encapsulated as different elementary streams. Therefore, left and right views are transmitted in different time slices or bursts. The link layer output MPEG-2 Transport Stream (TS) packets are passed to physical layer where the transmission signal is generated with a DVB-T modulator. After the transmission over a wireless channel, the receiver receives distorted signal and possibly erroneous TS packets are generated by the DVB-T modulator. The received stream is decoded using the section erasure method, i.e. the MPE-FEC frame is filled with contents of the error-free MPE and MPE-FEC sections and the empty bytes in the frame are marked as erasures, RS decoding is performed to reconstruct the lost data, and finally, the received and correctly reconstructed IP datagram are passed to the video client. IP datagram are handled in the stereo video streamer client and resulting NAL units are decoded with the stereo video decoder to generate right and left views. Finally, these views are combined with a special interleaving pattern to be displayed as 3D in the displayer. Figure 3-2 MPE-FEC frame structure. 3.2 UEP Approaches for Stereoscopic Video Transmission over DVB-H In [11], the necessity of using MPE-FEC for stereo video broadcasting over DVB-H was shown. In addition, significance of left and right views differs for simulcast and MVC coded videos. Unlike simulcast coding where left and right views have equal priorities due to independent coding; in case of MVC, left and right views possess different priorities. Since the right view is predicted from the left view, any error on the left view will directly affect the quality of the right view also causing a decrease on the overall quality. The straightforward way to transmit joint or simulcast coded stereo video is to use equal error protection (same MPE-FEC rate) for both left and right views. However, this technique does not Page 23 of 44

25 utilize the dependency of the views in case of MVC. The right view cannot be reconstructed regardless of how heavily it is protected or even received with no errors; if the left view is lost. Therefore use of Unequal Error Protection (UEP) between left and right views rather than Equal Error Protection (EEP) seems to be a better strategy for MVC. We evaluated the impact of MPE-FEC by evaluating different MPE-FEC code rate assignments unequally on left and right sequences. In order to achieve different code rates, we employed puncture and padding operations. MVC FEC Rate PSNR Video bit rate (Kbps) L R L R Joint L R Total NO-PRO NO FEC NO FEC 31,79 31,53 31,66 193,25 122,95 316,20 EEP-1 3/4 3/4 30,45 30,15 30,30 143,44 87,62 231,06 EEP-2 7/8 7/8 31,00 30,72 30,86 163,08 101,51 264,59 UEP-1 3/4 7/8 30,45 30,64 30,54 143,44 106,96 250,40 UEP-2 3/4 NO FEC 30,45 31,38 30,89 143,44 139,01 282,45 Simulcast L R L R Joint L R Total NO-PRO NO FEC NO FEC 30,45 30,32 30,38 143,44 143,43 286,87 EEP-1 3/4 3/4 29,14 29,02 29,08 104,68 103,27 207,95 EEP-2 7/8 7/8 29,79 29,67 29,73 123,51 122,43 245,94 Table 3-1 Error resilience strategies and related video coding values (Video Set 1) used in the simulations 3.3 Experimental Results The simulations are carried out using the following DVB-H transmission parameters: 16-QAM or QPSK, 1/2 or 2/3 convolutional code rate, 8-K OFDM mode, 1/4 guard interval, 8MHz bandwidth and 3 sec Delta-t. Transmission parameters and simulated channel models are provided in Table 3-2. Parameter Set Modulation FFT GI CR Channel Model 1 16-QAM 8K 1/4 1/2 TU6 with fd-max=24hz 2 16-QAM 8K 1/4 2/3 2 sets of VU, MR. PI 3 QPSK 8K 1/4 1/2 TU6 with fd-max=24hz 4 QPSK 8K 1/4 2/3 2 sets of VU, MR. PI Table 3-2 Transmission parameters and corresponding channel models used in simulations The simulations presented in this report are carried out using the stereo Horse sequence prepared by KUK Filmproduktion GmbH and which is available in the stereo video database of Mobile3DTV Project [22]. Experiments are repeated 80 times for TU6 channel model and 40 times for VU, MR and PI channel models both for Simulink model traces and WING-TV traces. The quality of received and decoded video is measured by computing Y-PSNR and SSIM of the views using original, uncompressed video as the reference. In all the figures, PSNR values are in db scale and calculated according to the following formulas where D l and D r Represent the mean squared error in left and right views respectively. Page 24 of 44

26 SSIM indices are computed using the following formula [21]. Where; µ x the average of original image; µ y the average of distorted image; σ 2 x the variance of original image; σ 2 y the variance of distorted image; cov xy the covariance of distorted image; c 1 = , c 2 = MVC CR 2/3 TS Bit Rates 16-QAM CR ½ L R Total L R Total NO-PRO EEP EEP UEP UEP Simulcast L R Total L R Total NO-PRO EEP EEP CR 2/3 QPSK CR ½ MVC L R Total L R Total NO-PRO EEP EEP UEP UEP Simulcast L R Total L R Total NO-PRO EEP EEP Table 3-3 TS bit rates of Video set 1 after encapsulated for different transmission schemes Table 3-1 summarizes different MPE-FEC code rate assignment schemes used in the simulations. For each scheme, the resultant video PSNR and bit rate values after the compression are presented, Final Transport Stream (TS) bitrates are provided in Table 3-3 for video set 1. For all cases, the video is encoded such that the resultant TS bit rate does not exceed the available bit Page 25 of 44

27 rate defined by the transmission parameters with 16-QAM modulation. For example, if a scheme uses stronger FEC protection for a view, then it either reduces the video bit rate or assigns weaker FEC protection for the other view for a fair comparison. During the encoding, we adjust the video bit rate by varying QP values. In the table, EEP-1 and EEP-2 correspond to applying equal error protection to both streams, UEP-1 and UEP-2 correspond to applying different error protection rates to both streams and NO-PRO corresponds to streaming without error protection (MPE-FEC functionality switched off). We simulate simulcast coding with only EEP as the left and right sequences are coded independently. Before we show the performance of UEP strategy, we first compare simulcast coding and MVC coding using the NO-PRO and EEP-1 schemes shown in Table 3-1. The average PSNR plots for left view, right view and joint case are shown in Figure 3-3, Figure 3-4 and Figure 3-5. As seen from the figures, the simulcast case is better only in a few very low channel SNR cases. This can be explained by the fact that in low channel SNR where MPE-FEC protection is not efficient, the frequent losses in left view also affects the right view for MVC case. However, as channel SNR increases, MVC case performs much better than simulcast case. Figure 3-3 PSNR comparisons of MVC and simulcast coding strategies for left sequence, Modulation: 16-QAM, Channel Model: TU6, Page 26 of 44

28 Figure 3-4 PSNR comparisons of MVC and simulcast coding strategies for right sequence, Modulation: 16-QAM, Channel Model: TU6 Figure 3-5 Joint PSNR comparisons of MVC and simulcast coding strategies for both sequences, Modulation: 16-QAM, Channel Model: TU6 Page 27 of 44

29 Next, we compare the different MPE-FEC rate allocation schemes shown in Table 3-1 for MVC coded video. The average PSNR figures for right view and joint case are shown in Figure 3-6 and Figure 3-7. First of all, the results show a clear improvement of quality when the received MPE- FEC data is used especially in low channel SNR cases together with almost no loss performance for high channel SNR values. When FEC rate increases, again a clear improvement on quality is seen in low channel SNR cases. Another observation from Figure 3-6 and Figure 3-7 is that, better quality can be achieved by UEP compared to EEP. This can be seen by comparing EEP-2 and UEP-2. In EEP-2 scheme, both views use a FEC rate of 7/8. In UEP-2 scheme, right view is not protected at all and left view is protected with a FEC rate of 3/4, which is stronger than the 7/8 of EEP-2. Therefore, UEP-2 case is almost equivalent to using parity bits of the right view in EEP-2 for stronger protection of the left view. The figures show that, UEP-2 achieves significant PSNR gains especially in low channel SNR cases. The comparison of EEP-1 and UEP-1 is different from EEP-2 and UEP-2 comparison. The reason is that both EEP-1 and UEP-1 use same FEC protection rate for left view but for right view, UEP-1 method uses the excessive bit rate earned from protecting right view weaker to encode video with a better quality. As a result, EEP-1 performs better in low channel SNR as it employs stronger protection and performs worse in good channel conditions as UEP-1 encodes video with a better quality. Figure 3-6 PSNR comparisons of MPE-FEC rate assignment strategies for right sequence, Coding: MVC, Modulation: 16-QAM, Channel Model: TU6 Page 28 of 44

30 Figure 3-7 Joint PSNR comparisons of MPE-FEC rate assignment strategies for both sequences, Coding: MVC, Modulation: 16-QAM, Channel Model: TU6 Apart from 16-QAM modulation, we also repeated the same experiments with QPSK modulation. Since QPSK results in half bit rate compared to 16-QAM, we adjust all the video bit rate parameters in Table 3-1 Error resilience strategies and related video coding values (Video Set 1) used in the simulationstable 3-1such that the video bitrates are halved as well. We show the results of Simulcast-MVC coding comparison and different MPE-FEC rate schemes for average joint PSNR case in Figure 3-8 and Figure 3-9. The results for QPSK modulation are in agreement with 16-QAM case, confirming our deductions. Comparing the results of QPSK and 16-QAM modulation, QPSK achieves higher PSNR values in very low channel SNR values. This is expected since QPSK is more robust to channel errors than 16-QAM. However, since the data bit rate is halved, maximum quality that can be achieved by QPSK is much lower than 16-QAM and as the channel conditions get better, 16-QAM performs significantly better. Page 29 of 44

31 Figure 3-8 Joint PSNR comparisons of MVC and simulcast coding strategies for both sequences, Modulation: QPSK, Channel Model: TU6 Page 30 of 44

32 Figure 3-9 Joint PSNR Comparisons of MPE-FEC rate assignment strategies for both sequences, Coding: MVC, Modulation: QPSK, Channel Model: TU6 In the following part, Figure 3-10, Figure 3-11 and Figure 3-12 demonstrates PSNR comparison of error resilience strategies for MVC coded videos, transmitted with 16-QAM, simulated on the new models of DVB-H, namely MR, PI and VU that are obtained from both Simulink Model traces and WING-TV traces. Afterwards, Figure 3-13, Figure 3-14 and Figure 3-15 demonstrates same experiments for Simulcast coded video. Afterwards, we tabulate the results of experiments on TU6 Channel Model in Table 3-4 and Table 3-5. The average distortions in Table 3-4 are given for PSNR values and the ones in Table 3-5 are given in SSIM metric. Page 31 of 44

33 Figure 3-10 Joint PSNR results of MVC coded video, transmitted with 16-QAM through Motorway Rural by Simulink Model (upper) and WING-TV Models Page 32 of 44

34 Figure 3-11 Joint PSNR results of MVC coded video, transmitted with 16-QAM through Pedestrian Indoor by Simulink Model (upper) and WING-TV Models Page 33 of 44

35 Figure 3-12 Joint PSNR results of MVC coded video, transmitted with 16-QAM through Vehicular Urban by Simulink Model (upper) and WING-TV Models Page 34 of 44

36 Figure 3-13 Joint PSNR results of Simulcast coded video, transmitted with 16-QAM through Motorway Rural by Simulink Model (upper) and WING-TV Models Page 35 of 44

37 Figure 3-14 Joint PSNR results of Simulcast coded video, transmitted with 16-QAM through Vehicular Urban by Simulink Model (upper) and WING-TV Models Page 36 of 44

38 Figure 3-15 Joint PSNR results of Simulcast coded video, transmitted with 16-QAM through Pedestrian Indoor by Simulink Model (upper) and WING-TV Models Page 37 of 44

39 Channel MVC 16QAM Simulcast 16QAM SNR No Pro EEP1 EEP2 UEP1 UEP2 No Pro EEP1 EEP Channel MVC QPSK Simulcast QPSK SNR No Pro EEP1 EEP2 UEP1 UEP2 No Pro EEP1 EEP Table 3-4 Simulation results in average PSNR for 16-QAM and QPSK modulations through TU6 Channel Page 38 of 44

40 Channel MVC 16QAM Simulcast 16QAM SNR No Pro EEP1 EEP2 UEP1 UEP2 No Pro EEP1 EEP Channel MVC QPSK Simulcast QPSK SNR No Pro EEP1 EEP2 UEP1 UEP2 No Pro EEP1 EEP Table 3-5 Simulation results in average SSIM for 16-QAM and QPSK modulations through TU6 Channel Another set of transmission experiments with assignments of NO-PRO, EEP and UEP strategies to videos of higher bitrates (Video Set 2) are presented in Table 3-6. This experiment covers the transmission of MVC coded videos protected with different FEC rates, over the TU6 Channel with maximum Doppler frequency offset of 24 Hz. Physical layer transmission parameters are from parameter set 1 of Table 3-2. The PSNR results for Joint, Left and Right sequences are provided in Figure 3-16,Figure 3-17 and Figure When we transmit the 3D video which is coded with higher bit rate using the same FEC assignments strategies, channel and transmission parameters; a higher TS bit rate is required as seen in Table 3-6. This is also equivalent to saying that we have a larger burst in this case. From the results, we see that higher PSNR values are reached for relatively good channel conditions. In Figure 3-16, the highest PSNR value is around 34 db whereas for the video set 1 (previous video set coded with lower bitrates) it was around 31.5 db which can be seen in Figure 3-7. But when we consider the PSNR values achieved in the worst channel condition, i.e. SNR 13 db; videos with lower bit rate seem to have a better performance having higher PSNR values. IP encapsulation requires the network layer packets to be less than the maximum transferrable unit. This causes the fragmentation of larger frames when they are passed to network layer. However when a NALU is fragmented while IP encapsulation, even one of the fragments is lost that NALU cannot be Page 39 of 44

41 recovered. This causes a larger error rate than for a case with many small NAL units having a total size equal to the one large NALU. Also there are many studies showing the improved performance of H.264 with NALU sizes smaller than maximum transfer unit [23]. Therefore we observe an expected decrease in PSNR values for the high bit rate video rate due to the fragmentation of large frames. MVC pro TS Bit Rate L R L R Total NO-PRO no fec no fec EEP-1 3/4 3/ EEP-2 7/8 7/ UEP-1 3/4 7/ UEP-2 3/4 no fec Table 3-6 Error resilience strategies and related TS bitrates for the videos coded with higher bitrates Figure 3-16 Joint PSNR comparisons of MPE-FEC rate assignment strategies for joint sequence of video set 2, Coding: MVC, Modulation: 16-QAM, Channel Model: TU6 Page 40 of 44

42 Figure 3-17 PSNR comparisons of MPE-FEC rate assignment strategies for left sequence of video set 2, Coding: MVC, Modulation: 16-QAM, Channel Model: TU6 Page 41 of 44

43 Figure 3-18 PSNR comparisons of MPE-FEC rate assignment strategies for right sequence of video set 2, Coding: MVC, Modulation: 16-QAM, Channel Model: TU6 Page 42 of 44

44 4 Conclusions In this report, we study the impact of MPE-FEC over the quality of the delivered stereo video for different channel models and different conditions of the same channel. As shown in the experimental results, although MPE-FEC provides the much needed data robustness for 3D video transmission in wireless channels, under very erroneous conditions it may fail. In order to overcome this, we use a-priori knowledge of the media to differentially protect data using FEC. High priority (left) video is well protected and low priority (right) video is less protected. The results show a clear improvement on the received video quality. These observations can be done based on the SIMULINK models since we can generate traces for a wide SNR range. The results obtained using the Wing-TV traces are compatible with the corresponding range of the SIMULINK models However, general conclusions cannot be derived using the available Wing-TV traces since they provide narrower SNR ranges from which the improvements of UEP cannot be shown. 5 Acknowledgements We would like to thank Dr. Ali Hazmi from the Department of Communication Engineering, Tampere University of Technology for providing simulated channel error traces based on Wing TV DVB-H channel models. Page 43 of 44

45 6 References [1] COST207, Digital Land Mobile Radio Communications (Final Report), Commission of the European Communities, Directorate General Telecommunications, Information Industries and Innovation, 1989, pp [2] Joint Technical Committee of Committee T1 RlP1.4 and T1A TR46.3.3iTR on Wireless Access, Draft Final Report on RF Channel Characterisation, Paper No, JTC(AIR)/ R4J,a n. 17, [3] CELTIC-WINGTV Project ( ) [4] ETSI TR v1.4.1, DVB BlueBook A092 Rev.3, Digital Video Broadcasting (DVB); DVB-H Implementation Guidelines [5] ETSI TS V1.3.1, DVB BlueBook A101 Rev.1, IP Datacast over DVB-H: Content Delivery Protocols (CDP) [6] ETSI draft TS V1.1.1, Digital Video Broadcasting (DVB); MPE-IFEC [7] Gómez-Barquero D., Gozálvez D., Cardona N., Application Layer FEC for Mobile TV Delivery in IP Datacast Over DVB-H Systems, IEEE Transactions on Broadcasting [8] ETSI EN V1.6.1, Digital Video Broadcasting (DVB);Framing structure, channel coding and modulation for digital terrestrial television [9] ETSI TM 3783 DVB BlueBook A115 - DVB Application Layer FEC Evaluations [10] ISO/IEC JTC1/SC29/WG11, Overview of 3D Video Coding," May Doc. N9784. [11] M. O. Bici, A. Aksay, A. Tikanmaki, A. Gotchev, and G. Bozdagi Akar, "Stereo Video Broadcasting Simulation for DVB-H," In NEM-Summit'08, [12] S. Cho, N. Hur, J. Kim, K. Yun, and S. Lee, "Carriage of 3D audio-visual services by T- DMB," Electronics and Telecommunications Research Institute, Republic of Korea, in Proc ICME, [13] E. Failli, "Digital land mobile radio," Final report of COST 207, [14] G. Faria, J. Henriksson, E. Stare, and P. Talmola, "DVB-H: Digital broadcast services to handheld devices," Proceedings of the IEEE, 94(1):194{209, [15] B. Furht and S. Ahson, "Handbook of Mobile Broadcasting: DVB-H, DMB, ISDB-T, and MediaFLO," Auerbach Publications, [16] "3dphone project" [17] P. Merkle, A. Smolic, K. Mueller, and T. Wiegand, "Comparative study of mvc prediction structures," Joint Video Team of ISO/IEC MPEG & ITU-T VCEG. [18] M. Oksanen, A. Tikanmaki, A. Gotchev, and I. Defee, "Delivery of 3D Video over DVB-H: Building the Channel," In NEM-Summit'08, [19] P. Pandit, A. Vetro, and Y. Chen, "Jmvm 3 software," ITU-T JVTV208, [20] A. Vetro, P. Pandit, H. Kimata, and A. Smolic, "Joint Draft 3.0 on Multiview Video Coding," Joint Video Team, Doc. JVT- W, [21] Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, "Image quality assessment: From error measurement to structural similarity," IEEE transactions on image processing 13(4):600{612, [22] "Mobile3dtv: Mobile 3DTV content delivery over DVB-H system" [23] J. Korhonen and P. Frossard, Error control for Video Streaming with Small Data Units, Proceedings of ICST Mobimedia, Page 44 of 44

46 Mobile 3DTV Content Delivery Optimization over DVB-H System MOBILE3DTV - Mobile 3DTV Content Delivery Optimization over DVB-H System - is a three-year project which started in January The project is partly funded by the European Union 7 th RTD Framework Programme in the context of the Information & Communication Technology (ICT) Cooperation Theme. The main objective of MOBILE3DTV is to demonstrate the viability of the new technology of mobile 3DTV. The project develops a technology demonstration system for the creation and coding of 3D video content, its delivery over DVB-H and display on a mobile device, equipped with an auto-stereoscopic display. The MOBILE3DTV consortium is formed by three universities, a public research institute and two SMEs from Finland, Germany, Turkey, and Bulgaria. Partners span diverse yet complementary expertise in the areas of 3D content creation and coding, error resilient transmission, user studies, visual quality enhancement and project management. For further information about the project, please visit Tuotekehitys Oy Tamlink Project coordinator FINLAND Tampereen Teknillinen Yliopisto Visual quality enhancement, Scientific coordinator FINLAND Fraunhofer Gesellschaft zur Förderung der angewandten Forschung e.v Stereo video content creation and coding GERMANY Technische Universität Ilmenau Design and execution of subjective tests GERMANY Middle East Technical University Error resilient transmission TURKEY MM Solutions Ltd. Design of prototype terminal device BULGARIA MOBILE3DTV project has received funding from the European Community s ICT programme in the context of the Seventh Framework Programme (FP7/ ) under grant agreement n This document reflects only the authors views and the Community or other project partners are not liable for any use that may be made of the information contained therein.