Digital Cinema Delivery using Frequency Multiplexed DVB T Signals

Digital Cinema Delivery using Frequency Multiplexed DVB T Signals Giuseppe Baruffa #1, Paolo Micanti #2, Fabrizio Frescura #3, Saverio Cacopardi #4 # Dept. of Electronic and Information Engineering, University of Perugia Via G. Duranti 93, 06125 Italy 1 baruffa@diei.unipg.it 2 paolo.micanti@diei.unipg.it 3 frescura@diei.unipg.it 4 cacopardi@diei.unipg.it Abstract Digital Cinema (DC) electronic distribution requires networks with high available bandwidth. In this paper, we propose a system that adopts a frequency multiplex of DVB-T signals to deliver JPEG 2000 video over wideband wireless channels. The objective of this paper is to present a transmission scheme that, by transmitting over DVB-T channels, provides content distributor and/or end users with the possibility to obtain a very high quality video stream. The source coded video can be downsampled to a video quality still at an high level, and then compressed with parameters reflecting the specifications for DC. The stream is thus split and encapsulated into MPEG-2 TS packets. A proper strategy of substream multiplexing and composition is presented, which may reduce the residual error effects and bandwidth requirements. An additional Forward Error Correction (FEC) layer using JPWL can be applied to further enhance transmission robustness. At the receiver side a proper demultiplexing and decompression is performed. Eventually, we will also present simulated results in terms of objective video quality vs. channel SNR or equivalent bit error rate. I. INTRODUCTION The Digital Cinema standard is widely adopted, and is going to replace analogue cinema; among its advantages, there is a very high visual quality, easier storage and transport, and higher manageability, due to the possibility of accessing digital content using transmission networks. The DCI consortium standardized the characteristics for DC [1], specifying 2K (2048 1080) and 4K (4096 2160) resolutions, XYZ color space, 12 bit/sample precision, 4:4:4 chroma subsampling, JPEG 2000 image compression. These parameters lead to a very high transmission bandwidth that cannot exceed 250 Mbps, which makes real time DCI streaming unfeasible on most wireless transmission networks. Recent works address the transmission of JPEG 2000 content on various networks (wired [2], [3] or wireless [4], such as WiMAX [5] or OFDM [6]); in this paper we present a strategy for both real time and off-line transmission of DC like streams on DVB T channels. DVB T has never been thoroughly investigated (as far as we know) as a possible transmission medium for JPEG 2000 compressed images or Digital Cinema streams: our objective is to examine an architecture and integration technique between DVB-T channels and JPEG 2000 codestreams, for transmission of real time and offline DC high definition video content. Existing literature reports some research to design a JPEG 2000 scalable video transmission system over DVB-H, providing large gains in terms of power savings for mobile receivers [7]. In our case, being energy consumption not an issue, we focus the interest on the layered scalability, which may lead to a great advantage in terms of visual quality. By decreasing the request on image parameters (e.g., using a different chroma subsampling, 4:2:2 or 4:2:0, and hence a different color space, YCbCr or YCxCz), we could lower the bitrate requirement and at the same time maintain a very high visual quality. It is interesting to note that such subsampling schemes are used in the stereoscopic version of the DCI standard [8], where two separated 2K video streams are multiplexed in the same package, which must conform to the 250 Mbps limitation. The typical setup of DC delivery over wireless channels is depicted in Fig. 2. For real time transmission, a typical bitrate interval of 80 100 Mbps could be achieved, which would require from two to four 8 MHz DVB T channels operating at high speed (i.e., 64QAM modulation, 7/8 coding, 1/32 guard interval). In the off-line case, it would be possible to use a single DVB-T channel, maintaining the original video quality, but also in this case a bitrate reduction and/or multi-channel transmission can be envisaged for download time saving purposes. Moreover, the DVB T standard offers a technique called Hierarchical Transmission [9], which allows for the transmission of two different transport streams at the same time; one is more Fig. 2. Transmission scenario

Fig. 1. Scheme of DC transmission using N m frequency multiplexed DVB-T signals robustly protected (e.g., low coding rates such as 1/2 or 2/3, and QPSK modulation) and may be used for the high priority multimedia data (HP), while the other exploits the lightest protection (e.g., coding rate 7/8 and 64QAM modulation, where only the 4 LSBs are used for this purpose) and is used for the low priority multimedia data (LP). It is interesting to note that this feature of the DVB-T standard is generally not used in commercial broadcasting, in spite of its promised advantages. To further refine our system, we should also take into account two main features that are required for full exploitation of this system: the transmission of audio and other metadata (e.g., subtitles) in addition to the video contents, and the encryption to prevent unauthorized access to the data. In the paper we are not going to investigate on the said points, being beyond our current scope. However, it would be straightforward to take into account the additional bit rate needed for audio delivery and the management of cryptographic operations for ciphering and deciphering the stream (e.g. using the industry-standard AES 128 bit algorithm). II. REFERENCE WORKS There is little existing work done on the specific argument of JPEG 2000 transmitted over DVB T/OFDM links. In [6] the authors adopt a method to embed FEC data into the last layer of a quality scalable codestream, thus keeping full backward compatibility with JPEG 2000 Part I codestreams and decoding architectures. Basically, they duplicate the data contained in the main header and in the first layer (being it the most significant to the image quality), compute the parity by means of a RS(255,239), and replace all these data in the last layer, which has a negligible impact in terms of PSNR performance when discarded at the receiver. However, they do not protect at all the intermediate quality layers. Differently, in [10], the authors do not use FEC to protect image data, but they envisage a prioritized transmission by loading multimedia bits on the OFDM carriers depending on their significance and importance in the codestream: the bitstream is reordered, and the more important data are transmitted over OFDM subchannels characterized by lower channel estimation error (thus, equalization and performance for those carriers is improved). However, they do not apply this method to JPEG 2000, but to a similar scalable codec (SPIHT). In [4], a product of Turbo and RS codes was used for transmitting standard definition images over a fading channel. They associate a larger amount of FEC data with packets that are more important, and the source datastream is protected by channel codes arranged as a product code. By looking at DC transmission itself, in previous works [5], [11] we have investigated on the feasibility of the transmission of DC over wireless channels (such as WiMAX) characterized by a static packet loss rate. By using JPWL (this standard [12] offers a straightforward extension to JPEG 2000 coding, conceived to deal with the transmission of codestreams over wireless channels) and a virtual pre-interleaving technique, the performance of this system was shown to be acceptable for the typical scenario of digital delivery. III. SCENARIO The scenario is that of electronic transmission of Digital Cinema streams using a wireless channel. Among the advantages of DVB-T we cite its high reliability and diffusion, and its high transmission range. This architecture could be used for transmitting DC contents to theaters where no high bandwidth wired channels are available. Also, this could be adapted to real time HD live events transmission. DC transmission to mobile users could also be feasible: many portable devices, such as notebooks and netbooks, PDAs, advanced mobile phones, have an integrated DVB T receiver. Such users could be allowed to receive a JPEG 2000 encoded single-channel DVB-T stream, and a simple software and firmware update should allow to display the relevant video stream: even if JPEG 2000 is a complex codec, it is expected that the use of its inherent resolution scalability as well as the advancement in computational power of mobile devices CPUs, will allow to decode the stream in real-time, even if at a lower resolution than the DC standard one. DVB-T offers some advantages: first of all, it is broadcasted, which would let the content provider reach a wide range of final users without additional costs or protocol burdens. The

Fig. 3. Higher and lower JPEG 2000 resolution and quality layers are conveyed via HP and LP services same goal is not currently trivial for WiMAX or other wireless transmission protocols. As a bonus, this standard is also widely adopted in a large number of countries. Last but not least, the upcoming enhancement of the terrestrial standard, DVB T2, will allow for a 30% increment in the available bitrate without bandwidth expansion, using more efficient modulation and channel coding schemes [13]. Thus, an investigation of the first generation system performance in this field will open the path to an extension of the currently envisaged architecture. IV. SYSTEM ARCHITECTURE The proposed system architecture is depicted in Fig. 1. Original uncompressed DC sequences, XYZ color space, 4:4:4 36 bits, 24 fps, are subsampled to 4:2:2 or 4:2:0 30 bits and converted to YCxCz (or YCbCr) color space. This operation decreases the required bitrate for transmission, while having a low impact on perceived video quality (luminance component, to which human visual system is more sensitive, is not subsampled). Each video frame is compressed using the JPEG 2000 standard; several substreams are created, either with resolutionscalable or quality-scalable intents. JPWL may be employed to further provide a light yet powerful protection, which is able to correct remaining error bursts (after physical layer decoding) thanks to its interleaving capabilities. Quality/resolution layers are split into N m streams, each corresponding to a single channel of the final frequency multiplex: the streams are then sent with DVB T, as shown in Fig. 3. The low quality, high priority service is required for a correct decoding of the compressed image, while users receiving the higher quality, low priority service can exploit the full visual quality. In the final deployed scenario, another important constraint should be considered, that is an added synchronization layer for broadcasting the stream to many users. A. Encapsulation strategy An important feature of this system is represented by the encapsulation strategy. In fact, JPEG 2000 codestreams can be atomically represented by a main header (conveying information such as resolutions, layers, data precision, and subsampling) and a number of data packets, of variable size and number (conveying the lossy compressed and entropically encoded image samples). Due to its repetitive nature, main header transmission may be avoided, with the video parameters being transmitted seldom or in the delivery setup phase (they are assumed not to change during the feature). Thus, only data packets need to be encapsulated in MPEG-2 Transport Stream (TS) packets, which have a fixed size of 188 bytes. Because of the size of the transport packet, it is important to add only a small overhead for management and synchronization purposes. Thus, beyond the mandatory MPEG-2 TS sync packet (0x47), which occupies the first position, we have used only three single-byte fields to signal codestream properties (Fig. 4): Codestream number, used to identify the sequence number of the frame being transmitted (this number will be re-used normally after ten seconds, at typical frame rates, thus avoiding collision problems in the short term); Packet number, used to identify the syntactical position of the current packet in the codestream. This number may be used to univocally identify the scope of this packet (i.e., a quality enhancement or resolution enhancement packet). For the present work, we have used a combination of resolutions, layers, and codeblock sizes that keep the packet number below 255; Codestream fragment, used to identify the fragment of the packet being transmitted and if this is the last one. The maximum size of a codestream packet is thus of about 184*255=46 kbytes, which gives a maximum codestream size of 184*255*255=11.4 Mbytes (well beyond the 250 Mbps bitrate limitation). At the receiver side, a corresponding decapsulator must operate to recompose all of the JPEG 2000 packet fragments in the correct order, by avoiding frame mismatches and also trying to minimize the impact of uncorrected error events (for example, discarding out-of-order sequence numbers or invalid termination codes). After that, the JPWL decoder corrects the remaining errors due to missing received packets or long error bursts. Eventually, the received stream may be saved for later display or immediately rendered on the screen after source decoding (typical application is that of live events Fig. 4. Encapsulation strategy of JPEG 2000 data packets into MPEG-2 TS packets

broadcasting). The problem of achieving correct synchronization among separate channels may be solved with a correct buffering strategy, with the received codestream fragments being saved in a temporary pool from where they are extracted for building the entire frame, which will be decoded and displayed when all the fragments have been received. The buffering delay may be large and it does not represent a problem, in this case, since we do not require an interaction to be set up between the transmitter and the receivers. Moreover, the DVB-T SFN operation mode already embeds a particular framing method (i.e., megaframes) that allows to achieve a very fine synchronization at the moment of transmitting the same transport stream over multiple different transmitters. B. Layering and resolution scaling vs. throughput There are several methods which may be adopted for preparing JPEG 2000 packets, according to predefined strategies in quality layering and resolution analysis. For instance, in case of 4K content delivery, the HP service may be used for conveying the basic 2K content, whereas the LP one carries the packets necessary to reach full resolution. Differently, two quality layers may be envisaged, where one is used to provide a low resolution, low quality version of the data, and the remaining ones will carry the rest. Table I shows two possible 8 MHz channel DVB-T and JPWL parameter configurations (C1 and C2) for achieving a throughput of 100 and 80 Mbit/s, respectively, which may be sufficient for delivering DC with good quality. In the first case, configuration C1 gives a good throughput, at the expense of reduced data protection: this may be the case when the geographical area to be covered with DVB-T delivery is small (e.g. urban). On the other side, configuration C2 gives a more rugged environment for data transmission, which may well fit to a mixed rural/urban coverage area. V. SIMULATION RESULTS The performance results can be given in terms of the received and decompressed video PSNR vs. the (C/N) ratio experienced on the transmission channel. We provide some results obtained by considering the binary symmetric equivalent channel (BSEC) of the DVB-T transmission system; the mapping between (C/N) ratios and the equivalent BER may be found easily in the literature (for example, in [14] are reported the BERs after Viterbi decoding, which may be mapped to post-reed-solomon BERs quite easily). For the moment we have considered 2K resolution DC images in RGB colorspace, without subsampling, compressed with JPWL at the rates specified in configurations C1 and C2, without quality layer scalability. The corresponding MPEG- 2 TS packets have been transmitted through the BSEC with different bit error probabilities P e, and the corresponding average PSNR has been computed. The results are reported in Tab. 2 and graphically presented in Fig. 5. The more rugged configuration, C2, enables correct reception of the video up to a BER of 1 10 3 on the transport TABLE I PARAMETER CONFIGURATION SETTINGS FOR 8MHZ DVB-T CHANNELS (BITRATES ARE EXPRESSED IN MBPS) C1 C2 Service HP LP HP LP Mode 2k 2k Prefix 1/32 1/16 Modulation 64QAM 64QAM DVB-T FEC 3/4 7/8 1/2 5/6 Bit rate (single) 9.048 21.112 5.855 19.516 Nm 4 4 Bit rate (mux) 36.192 84.448 23.420 78.064 JPWL FEC RS(37,32) RS(40,32) Throughput (JPWL) 120.640 101.484 Throughput (net data) 104.337 81.187 stream, whereas configuration C1 fully works only up to 5 10 5. The switchover point between the two configurations is found at a BER of 2 10 4. The simulated error rates are typical of the steep slope of the BER curves obtained after RS decoding in DVB-T systems [14]. These results show that the use of an additional FEC layer (JPWL) enables a wider operational range of the system, at the expense of a slight decrease in objective video quality, thus it is suggested to always use it. VI. CONCLUSION In this paper we have presented a method for transporting JPEG 2000 compressed video of Digital Cinema over multiple DVB-T links. The achieved bitrates and the reliability of DVB- T are sufficient to deploy a delivery network for feeding theaters that are not easily reached by high-speed wired links. The simulation results show that, in practical cases, DVB-T FEC by itself is not able to provide a good video quality; thus, an additional layer of JPWL FEC and interleaving has been added to obtain good figures of merit. PSNR (db) 35 30 25 20 15 10 5 10 2 10 3 10 4 P e on BSEC 10 5 C1 w. JPWL C1 w/o JPWL C2 w. JPWL C2 w/o JPWL 10 6 10 7 Fig. 5. Simulation results on the BSEC for the two considered configurations

REFERENCES [1] Digital Cinema System Specification, Digital Cinema Initiatives, LLC Std. V1.1, 2007. [Online]. Available: http://www.dcimovies.com [2] T. Shimizua, D. Shiraia, H. Takahashia, T. Murookaa, and K. Obanaa, International real-time streaming of 4K digital cinema, Future Generation Computer Systems, vol. 22, no. 8, pp. 929 939, Oct. 2006. [3] E. Itakura, S. Futemma, G. Wang, and K. Yamane, JPEG2000 based real-time scalable video communication system over the Internet, in Proc. IEEE Consumer Communications & Networking Conf., LasVegas, NV, Jan. 2005, pp. 539 543. [4] N. Thomos, N. V. Boulgouris, and M. G. Strintzis, Optimized transmission of JPEG2000 streams over wireless channels, IEEE Trans. Image Process., vol. 15, no. 1, pp. 54 67, Jan. 2006. [5] G. Baruffa, P. Micanti, and F. Frescura, Error correction and concealment for JPEG 2000 video transmitted over wireless networks, in Proc. Int. Symp. Wireless Communication Systems, Trondheim, Norway, Oct. 2007, pp. 757 761. [6] K. Munadi, M. Kurosaki, K. Nishikawa, and H. Kiya, Error protection for JPEG2000 encoded images and its evaluation over OFDM channel, in Proc. Int. Symp. Circuits and Systems, Bangkok, Thailand, May 2003, pp. 432 435. [7] E. Belyaev, T. Koski, J. Paavola, A. Turlikov, and A. Ukhanova, Adaptive power saving on the receiver side in digital video broadcasting systems based on progressive video codecs, in Proc. 11th Int. Symp. on Wireless Personal Multimedia Communications, Lapland, Finland, Sep. 2008. [8] Stereoscopic Digital Cinema Addendum, Digital Cinema Initiatives, LLC Std. V1.0, 2007. [Online]. Available: http://www.dcimovies.com [9] Digital Video Broadcasting (DVB); Framing structure, channel coding and modulation for digital terrestrial television, DVB Project Std. ETSI EN 300 744 V1.5.1 (2004-11), 2004. [Online]. Available: http://www.etsi.org [10] C. Pandana, Y. Sun, and K. J. R. Liu, Channel aware unequal error protection for image transmission over broadband wireless lan, in Proc. Int. Conf. Image Process., Barcelona, Spain, Sep. 2003, pp. 93 96. [11] G. Baruffa, P. Micanti, and F. Frescura, Unequal error protection and interleaving for wireless transmission of JPEG 2000 video, IEEE Trans. Image Process., vol. 18, no. 2, pp. 346 356, Feb. 2009. [12] JPEG 2000 image coding system - Part 11: Wireless JPEG 2000, ISO/IEC JTC1/SC29/WG1 Std. Committee Draft version 0.1, 2004. [13] Frame structure, channel coding, and modulation for a second generation digital terrestrial television broadcasting system (DVB-T2), DVB Project Std. DVB document A122, 2008. [Online]. Available: http://www.dvb.org [14] J. M. Lago and F. Pérez-González, Analytical bounds on the error performance of the DVB-T system in time-invariant channels, in Proc. Int. Conf. Communication, New Orleans, LA, Jun. 2000, pp. 651 655.