HDTV over UWB: wireless video streaming trials and quality ofservice analysisDomenico Porcino (1), Bram van der Wal (2) , Ying Zhao (3)(1) Philips Research, UK ([email protected])(2) Philips Consumer Electronics, NL ( Department of Electronics Engineering, Tsinghua University, China ([email protected])ABSTRACTThe PULSERS EC Integrated Project investigated and developed leading wireless systems for highspeed short-range video transmissions. This paper presents results obtained with the PULSERSCommunications Platform 2, an hardware testbed which was specified, designed and implemented toverify the feasibility of a high quality cable-less video transmission system, a wireless version of theDigital Video Interface (DVI). Our experiments analysed the use of a cascaded compression and highspeed wireless link to transmit high quality HDTV video wirelessly at short distances (2-10m).The paper addresses the reasons for our architectural choice, comparing the video codecs and differentwireless solutions both in terms of achieved throughput and perceived mean opinion score registered insubjective testing. The experimental PULSERS testbed and the corresponding field results prove thathigh quality and high resolution short-distance wireless UWB video links are feasible today.1. Introduction: Wireless DVI System ConceptThe Ultrawideband (UWB) technology is one of the most innovative and promising wireless means toenable cable-less transmission of high-speed data between close-by nodes. The IST Sixth FrameworkIntegrated Project PULSERS ([1]) has investigated and developed UWB prototype hardware to verifythe real-life feasibility of some of the most ambitious UWB targets and measure their correspondingquality of service for high-quality wireless video transmission.Our study and development of the PULSERS Communication Platform 2 concentrated on Very HighData Rate (VHDR) systems in which the end-to-end target wireless troughouput data from atransmission point to a receiver is in excess of 1500 Mbit/s. The goal of our research was in fact thedemonstration of the feasibility of a wireless version of a Digital Video Interface (DVI) protocol.The DVI specifications [2] define a digital interface for use between a computing device and a displaydevice. These specifications were developed by an industrial consortium, the Digital Display WorkingGroup (DDWG), in the late 1990s. This interface allows transmission of bidirectional data between twoelectronic devices and is typically used in video transfer connections (flat panels LCD display). Thesimple and low-cost DVI connectors provide very good performance and allow system developersbandwidth throughputs over 1.5 Gbits/s (i.e. over 187 Mbytes/s). The specification enablesmanufacturers to implement a complete transmission and interconnect solutions. DVI has been designedto overcome one of the biggest problems of the analog Video Graphic Array (VGA) transmissionprotocol. The VGA interface, widespread in PC applications starting from 1987, had in fact beendesigned for analog signaling as typically used in old graphic applications. Modern devices, such asliquid crystal displays and plasma screens and PC graphic cards, are designed with internal digitalsignals and converting those to analog VGA for transmission over cable and then reconverting back todigital in the screen adds both to the cost and to the loss of quality in the final link budget. The necessarydigital to analog conversions at source (PC) side and digital to analog at the receiver (monitor) side,

which always have a finite precision, are elements which are not anymore necessary with the use of adigital interconnect protocol and should be avoided.DVI simply allows digital signals to be transmitted directly over cable without any conversion. And theDVI standard also includes the possibility to drive analog signals using the VGA standard for backcompatibility. Due to these properties, the DVI standard (and its HDMI evolution) seems to be the idealcandidate for taking over the future high-quality video connections. A wireless version of DVI, althoughchallenging, would help filling a product need in current top of the range electronic devices. Theanalysis of the feasibility of a cable-less DVI standard starts with considerations on the necessarythroughput. Commercial devices employing DVI connectors (projectors, TVs, camcorders, gameconsoles) support a variety of video resolutions (from VGA to QXGA), as shown in Table 1. Asexpected the biggest issues lie with the highest quality video systems (SXVGA and above), whichtransmit raw data in excess of 1.5 Gbit/s.CharacteristicVGAHorizontal pixel count640Vertical pixel count480Total pixel307200Total pixels @ 32bit color9830400Mbit/s at 25 frames/s - 32bit color245.8Mbit/s at 25 frames/s - 32bit color (25%blanking overheads)307Mbit/s at 25 frames/s - 32bit color(blanking DVI 8/10 encoding)369SVGAXVGA SXVGA UXGA 536480000786432 1310720 1920000 2073600 314572815360000 25165824 41943040 61440000 66355200 3111920207431465769441573230424883775Table 1: Data Throughput requirements for a Wireless DVI system at different resolutionsThe requirements and applications analysis presented in [3] lead to the definition of an innovative radioarchitecture in which a first stage of data (video) compression is cascaded to a very fast radio link inorder to keep the payload over air as low as possible, while satisfying the high quality videorequirements.CompressionEngineDataINUWB RXUWB TX3ComDecompressionEngine3ComDataOUTFigure 1: Schematic representation of the PULSERS architecture for wireless DVI2. Video Compression Choice and ComparisonWhile the principle of cascading a fast compression engine and a fast radio interface might be easy tounderstand, the challenges associated in the development/choice of an appropriatecompression/decompression (codec) for such case are many. In the context of this work we aimed toadopt the most suitable commercially available solution that would enable a good quality wirelesstransmission. Our goals were the highest possible payload with the lowest wireless throughput while

maintaining a good subjective quality of the resulting received image even in the case of heavyinterference, as might happen in wireless transmissions (blockages, temporary external interference etc).The theory of video coding presents us two distinct categories of video codecs: Lossless algorithms based on the known or guessed entropy of the content to be transmittedand on adequate generally small compression ratios and Lossy algorithms, which make use of more advanced techniques to achieve higher compressionratios, at the cost of some data loss in the original content.In order to estimate the visual quality of lossy codecs, we introduce two potential objective indicators ofthe quality of the received video signals: the Mean Square Error (MSE) between an original and acompressed image and the Peak Signal to Noise Ratio (PSNR) of the compressed image. The MSEbetween an uncompressed monochrome m x n image U and its compressed form C can be defined as:MSE 1 m 1 n 12U (i, j ) C (i, j ) m n i 0 j 0(1)In the case of colour images with a Red Green Blue (RGB) representation the MSE is calculated over thethree separate (R, G and B) colour channels as:MSE 1 1 m 1 n 11 m 1 n 11 m 1 n 1222[ U R (i, j ) C R (i, j ) U G (i, j ) C G (i, j ) U B (i, j ) C B (i, j ) ]m n 3 i 0 j 03 i 0 j 03 i 0 j 0This indication of the relative difference between the two images is the basis for the PSNR parameter:MAX U2MAX UPSNR 10 log10 () 20 log10 ()(2)MSEMSEwhere MAXU refers to the maximum pixel value of the uncompressed original image. With a linear pulsecoded modulation that employs k bits per sample, MAXU equates to (2k-1). For example in typical pixelsrepresentations of 8 bits per sample, MAXU is 255.Compression typePHY bandwidth (in Mbps) needed for visually lossless videoPicture qualityvisually lossless (PSNR 40dB) QXGA HDTVCompressionRatio /CRM-JPEG 743229517710869Table 2: Compression types and PHY bandwidth neededBesides the MSE and PSNR, other objective quality indicators have been proposed in the literature inrecent years. One example of an objective method successfully tested for measuring quality of video

tests is the Structural Similarity Index Measure (SSIM) [4]. This latest metric compares local pattersof pixel intensities after these have been normalised for luminance and contrast, allowing a moreprecise assessment of the perceptual errors seen by the human eyes in a disturbed video transmission.The aim of our work is to evaluate very high quality solutions, so we decided to set our specificationstowards a video codec with quasi-lossless properties and set a reference point at a PSNR of 40 dB(i.e. at a level where visually impeding errors are unlikely). Using this picture quality reference wecompared MPEG codecs (MPEG-2 and MPEG 4) with M-JPEG, M-JPEG 2000 and also very recentH.264, Windows Media Video 9 and DivX codecs, as shown in Table 2.D em onstrator W ireless D VID VI inCom puter 1D VI toHD-SDIconverterHDTVvideo intoH D-SD IY busVideosignalsplitterM -JPEG2000encoderClock data busCbC r busM -JPEG2000encoderY busM -JPEG2000decoderPCI busHostC PUM em orybusRAM disk withstoredM -JPEG2000contentG B ETHPCIR adioplatformPCI busDVI outHD-SDIto D VIconverterHD TVvideo out ofHD -SD IC om puter 2Videosignalm ergerCbC r busClock data busM -JPEG2000decoderPCI busHostC PUM em orybusRAM disk withreceivedM -JPEG2000contentR adioplatformG B ETHPCIPCI busFigure 2: Final configuration of the PULSERS Communications Platform 2 – VHDR transmissionsThe minimum video resolution we intended to transmit wirelessly was HDTV, but the data payloadthat could be sent over air was limited by the physical layer characteristics of the underlying radiotechnology, so a strong compression was needed anyway to support our video streaming.Comparing different types of codecs we noticed that Motion-JPEG2000 was the most robust of thevideo compression techniques. With built-in resynchronisation, error detection and concealmentmechanisms, Motion-JPEG2000 can tolerate bursty and single-bit errors and a fundamental featurefor the wireless transmission was the fact that errors did not propagate to successive sequences as inthe case of MPEG-2 and do not appear in blocks. Real-time symmetric (compression anddecompression) dedicated hardware exists on the market for MJPEG 2000 and represents an optimalchoice for high-resolution wireless video transfers.Our final system architecture for the wireless transmission of high quality (HDTV and above) videostreams is represented in Figure 2. The key elements in our testbed were therefore the highperformance (MJPEG 2000) video codec and the wireless (UWB) transceiver.

3. Wireless Video Streaming TestsAfter the definition of the system architecture and the practical realisation of the hardware prototypeswe set up a testing campaign to verify the quality of service achievable by our wireless videostreaming testbed. The tests started with the selection of a set of video sequences at HDTVresolutions. The video content source we used was formed by uncompressed (raw) YUV sequences(1920x1080 pixels), part of the Philips portfolio of videos inside the Innovation Lab of Philips. Theyare also used for HDTV plasma and LCD experimental panels quality checks.Each video frame is saved in uncompressed YUV format and putting a lot of them in one containerfile (DPX for example) results in a video sequence of uncompressed HDTV material. The sequencesused come from movies and sport events, as the PULSERS wireless DVI setup will mostly be usedin a home environment.We used 2 different video sources:Video Source n.1:Philips HDTV Promotion video showing recent movie scenes and sport event (47 seconds) Shows a mix of very fast changing action scenes as well as slower moving scenes Detailed explosions, fast moving objects and panoramic overviews with lots ofwarms colours and flashing lights.Figure 3 Sample screenshots from video sequence n.1Video Source n.2:HDTV Air show (32 seconds) Shows a mix of very fast changing action scenes as well as slower moving scenes Detailed zoom-in of aircraft and people faces, detailed clouds

Figure 4 Sample screenshots from video sequence n.2In order to test the different possible options and resolution we pre-processed the video sequencesand pre-recorded different files representing the video sequences at different resolutions and withdifferent compression ratios.4. Streaming Quality Tests ResultsWe were able to perform both a physical layer test related to the wireless part of the high qualityvideo streaming platform (effective throughput) and a full quality of service analysis if the end-toend gigabit video link. The main results are presented in the following paragraphs.Initial reference tests – effective PHY throughputIn the initial (reference setting) wireless video testing set-up we connected the transmitting streamingPC to different potential physical layer wireless transceivers interfaces, among which commercial802.11g equipment, 802.11 Pre-N equipment, Ultrawaves UWB transceivers, free space opticaltransceivers. We also set a reference point with a wired Gigabit Ethernet IP connection.The first metric we selected for our video tests was the effective throughput, which combinesthroughput with coverage area of the wireless network. All of our platforms were tested on theirthroughput in Mbps over distances from 1, 2, 3, 4, 6, 8, 11 and 15 meters. We could measure theexact data rates of the platforms that we used for the compressed scenario using the software packageIPerf. IPerf was originally developed as a modern alternative for measuring TCP and UDPbandwidth performance over an IP network. It is therefore a good tool to measure the effectivethroughput of a wireless network, as in the PULSERS test configuration. IPerf is also a tool tomeasure maximum TCP bandwidth, and allows the fine tuning of parameters connected to the UDPprotocol. Iperf reports in an easy format bandwidth, delay jitter, datagram loss.

Figure 5: Initial wireless video testing set-upThroughput rates at measured ring distancesWired LAN Gigabit Ethernet10000Throughput in MbpsWLAN 802.11g1000WLAN 802.11 Pre-N100ULTRAWAVES UWB (1stgen.) platformFree space optical link(uncompressed)10PULSERS UWB (2nd ters11meters15meters1DistanceFigure 6: Effective (uncompressed) throughput from different physical layers for the PULSERS VHDR PlatformFinal tests – video sequences QoSFor the final tests we placed the receiving platform on a stationary table below a flat screen highquality Philips TV. The transmitter was moved on a trolley along a straight line at constant spacingaway from the receiver, as shown in Figure 7.

Figure 7 PULSERS VHDR: TX (bottom in the picture) and RX (top) components 6 meters apart.As explained in the Chapter 2, different metrics exist and can help in measuring the quality of avideo signal. What has been found however is that the measured subjective quality of a videosequence is not always perfectly correlated to the objective inverse mean square error or to the peaksignal to noise ratio. Therefore in order to evaluate the quality of our wireless video streamingsolution we called in a panel of experts that scored each particular sequence seen on screen.Having registered good and stable connectivity results in the throughput of the PULSERS UWBplatforms up to 5 meters, we were expecting good QoS in the video sequences transmissions as well.But the subjective QoS tests at this distance were very disappointing. We therefore decided to reducethe test distance between transmitter antenna and receiver antenna to 1,5 meter, which was themaximum distance at which we could have error-free MPEG-2 transmissions. The MotionJPEG2000 content was stable up to about 3 meter, before noticeable blurring artefacts came on thescreen. But for the sake of comparison we had to keep the transmitter-receiver distance fixed to 1.5meter for all different physical layers. This 1,5-meter distance between transmitter and receiver alsorepresents a typical cable replacement distance.The first set of tests were performed in ideal conditions, without interference of other wireless links.The second set of tests were performed in conditions where we interfered the wireless links byplacing obstructive objects (person, flowers and cabinet) between transmitter-receiver. To addemulated errors in the Wired LAN (reference) setup we used a utility that overloaded the traffic overthe Wired LAN link called ‘Network Traffic Emulator’ from Nsasoft.

We compressed our Philips promotion video from raw video into H.264, MPEG-2 and MotionJPEG2000 formats. We made compressed content that would be able to run on the WLAN MIMOpre-N, Wired LAN and PULSERS UWB platforms.In the table below you see the description of bit rates used for each type of HDTV 1080i compressedvideo on the different platforms, which were evaluated. Note that we were not able to reproducevisual lossless video (PSNR 40 dB) for the WLAN MIMO 802.11 pre-N platform. In that case wewere bound to the hardware platform’s maximum throughput of 42 Mbit/s, which was short ofachieving a full PSNR of 40 dB.Final subjectiveQoS test - 1080i content descriptionPSNR 40dB (except for MIMO)MPEG-2H.264M-JP2000Mbit/sMbit/sMbit/sWired Gigabit LAN8050150WLAN MIMO pre-N4040408050150PlatformsPULSERS UWBTable 3 Final subjective QoS test - content descriptionThe marks from 1-10 were assigned exclusively for the subjective visual quality of the video on thereceiving display. Effects like skipping of frames (all), MPEG artefacts (MPEG-2 and H.264),motion blurring (Motion-JPEG2000) and complete black screens (FS optical link) in the sequencesaffected the subjective impressions of the testers. The tested video resolutions were the ones allowedby the related platforms. Only Gigabit Ethernet, PULSERS UWB platform and Free-Space opticallinks were able to transmit sequences at high quality resolutions (HDTV). The other (compressed)links were limited by the underlying radio technology.QoS marksQoS subjective tests without interference / ressedWired GigabitLANWLAN MIMOpre-NPULSERSUWBFS optical linkPlatformsFigure 8 Final PULSERS MOS in LOS at 1,5m distance – comparison of different PY and compressions –LOSconditionsWhen our subjective group took place in front of the display for the analysis of perceived quality forthe HDTV 1080i content, they were asked to give marks for the visual quality of the different videocompression types used and the different platforms used. The results were then averages and formed

the so callled Mean Opinion Score (MOS), the basis for comparison of the quality of service offeredby our cascaded compression and wireless transmission HDTV link.As we can see from Figure 8, our evaluators perceived the uncompressed Free-space optical link asthe absolute best, as expected. Close to this platform mark we see that the M-JPEG2000 and H.264compressed video on the PULSERS UWB and Wired LAN platforms are perceived almost with thesame quality as the uncompressed free space optical link. M-JPEG2000, H.264 and MPEG-2 wereperceived lower on the WLAN MIMO system, confirming the fact that this platform was not capableof supporting high quality/low compressed video data, compared to the Wired LAN and PULSERSUWB platform.When the QoS tests in LOS conditions were completed, the test group sit in front of the display againand they were asked to give new marks for the visual quality of the different HDTV 1080i videocompression types. This time the radio link was not in full line of sight, but was disturbed byobstructions and interference in the radio path as introduced with hand waving and walking troughthe link as it might happen in normal domestic situations. The results for these tests in Non Line ofSight (NLOS) obstructed conditions are mentioned in the Figure 9.QoS marksQoS subjective tests with interference / ssedWiredWLAN MIMOGigabit LANpre-NPULSERSUWBFS opticallinkPlatformsFigure 9 Final PULSERS MOS in LOS at 1,5m distance distance – comparison of different PY and compressions– with obstructionsIn the case of severe interference we can notice how the free space optical link was not judged asideal anymore. This was because it showed black frames as soon as the line of sight path betweentransmitter and receiver was blocked and this is a noticeable and clearly unacceptable behaviour.When we take a look at the M-JPEG2000 ratings we see that people prefer this compression methodinstead of MPEG-2 and H.264 on all platforms. This can be explained because of the different visualerrors appearing on the screen and noticed by our test group. People found much more annoying tosee the MPEG block artefacts coming on the screen and blocking a lot of visual information, ratherthan the blurred errors coming of the M-JPEG2000 final visual compressed videos (see Figure 10).

Figure 10: Different types of errors: blocky (left) for the MPEG-2 and blurred (right) for the M-JPEG2000 codecEven in very badly obstructed conditions, our test people were still able to see the content of theMJPEG 2000 video and only in some places in the screen the image was a bit blurrier, which isperceived as acceptable given the conditions.We also registered a noticeable difference between the ratings for M-JPEG2000 and those for H.264in the UWB PULSERS platform. This could be due to the fact that high bandwidth M-JPEG2000compared to MPEG-2 and H.264 is more resilient to block errors, with an efficient use of entropyencoding and frame by frame compression with no memory that could cause in the H.264propagation of errors.5. ConclusionsThe IST Sixth Framework Integrated project PULSERS has specified and developed an innovativeexperimental hardware architecture to allow high quality wireless video transmission and to helpverify the feasibility of a cable-less replacement of an advanced wired video transmission protocol,known as Digital Video Interface (DVI).The cascaded combination of fast compression (based on M-JPEG 2000, MPEG-2 and H.264techniques) and fast UWB radio links has been analysed and compared with reference fast wiredlinks (Gigabit Ethernet) and wireless Optical Links. The quality of the resulting link has beenmeasured both in terms of throughput and subjective perceived quality (Mean Opinion Scoreanalysis).From our experiments we can conclude that: For low quality (VGA-type) video, all the analysed radio platforms (802.11g, 802.11N,UWB) in conjunction with a hardware compression engine can deliver reasonably long radiolinks (typically over 10m), even if the subjective quality of the received video is notexcellent. For high-resolution videos (HDTV or above) only free-space optical links and the UWBPULSERS platform have been able to match the quality of wired solutions. The subjectivetesting confirmed that the quality of wireless videos is not yet exceptional in all conditions,but is already perceived as good for short (1.5m) distances.

The wireless DVI system implemented and tested in PULSERS during 2005 was top of the class andwas a good testbed to demonstrate feasibility of what could become a killer application of the future.Future work should now start also addressing possibilities for lower cost and lower complexitycompression techniques which could be coupled with the faster UWB platforms expected on themarket in the next few years.References[1] Sixth Framework Integrated Project PULSERS, IST FP6 506897,[2] Digital Display Working Group, “Digital Visual Interface DVI”, 02 April 1999[3] Domenico Porcino (Philips Research, UK), Bram van der Wal (Philips Consumer Electronics,NL), Ying Zhao (Department of Electronics Engineering, Tsinghua University, China), "Asimple architecture for a Wireless DVI system", IEEE International Conference on UltraWideband 2005 (ICU 2005), Zurich, Switzerland, September 5-8, 2005[4] Zhou Wang, Alan Bovik, Hamid Sheikh, Eero Simoncelli, “Image Quality Assessment: FromError Visibility to Structural Similarity”, IEEE Transactions on Image Processing, Vol 13, N4,April 2004

HD-SDI Host CPU PCI bus PCI bus RAM disk with stored M-JPEG2000 content Memory bus RAM disk with received M-JPEG2000 content M-JPEG 2000 decoder M-JPEG 2000 decoder Video signal merger Y bus CbCr bus Clock data bus Host CPU PCI bus PCI bus Memory bus HDTV video out of HD-SDI HD-SDI to DVI convert