the best of iet and ibc 2014 · the best of iet and ibc, 2014, vol. 6, pp. 1–2 2 & the...

The Best of IET and IBC 2014

INSIDE Papers and articles on electronic media technology from IBC2014 presented with selected papers from the IETrsquos flagship publication Electronics Letters

Volume 6 | ISSN 2041-5923

2014

Published by the Institution of Engineering and Technology wwwtheietorgjournals

Journals

IET IBC journal cover 2014indd 1 16072014 0909

Find out more about the DTG and see demonstrations from our winners in

Hall 5A09

wwwdtgorguk DigitalTVGroup

DTG

television

innovation

awards

Partnering UK SMEs for a future of TV technology showcase

For a second year Digital TV Group UKTIKTN and the IET have given four SMEsexhibition space at IBC to promote

UK innovation

Come and see the winning innovations that Come and see the winning innovations that will guide the future of TV including

-Viewer facial recognition-Sound triggered content delivery

-Big data analytics-Instantaneous video delivery to social media

DTG Television Innovation Award Winners 2014

Introduction 1

Editorial 3

Visions of Future Broadcasting at IBC2014

Selected content from IBC2014

DVB-T2 LITE exploiting broadcast HDTV networks for services to mobile receivers

G Alberico A Bertella S Ripamonti and M Tabone 5

Interview ndash Gino Alberico

An interview with the author of the paper selected as best young professional contribution for 2014 12

Computational imaging for stop-motion animated video productions Frederik Zilly Matthias Ziegler Joachim Keinert

M Schoberl and S Foessel 14

Interview ndash Frauhofer Institute for Integrated Circuits 20

THE ART OF BETTER PIXELS

DG Brooks 22

Source-timed SDN video switching of packetized video

TG Edwards 30

Live event experiences - interactive UHDTV on mobile devices

OA Niamut GA Thomas E Thomas R van Brandenburg L DrsquoAcunto R Gregory-Clarke 38

High frame-rate television sampling theory the human visual system and why the Nyquist-Shannon theorem

doesnrsquot apply

KC Noland 44

Development of 8K-UHDTV system with widecolour gamut and 120-Hz frame frequency

T Soeno T Yasue K Kitamura K Masaoka Y Kusakabe H Shimamoto T Yamashita Y Nishida M Sugawara 51

HEVC subjective video quality test results

TK Tan M Mrak R Weerakkody N Ramzan V Baroncini GJ Sullivan J-R Ohm KD McCann 59

Selected content from the IET

Introduction to Electronics Letters 65

Coherent THz communication at 200 GHz using a frequency comb UTC-PD and electronic detection

G Ducournau Y Yoshimizu S Hisatake F Pavanello E Peytavit M Zaknoune T Nagatsuma and J-F Lampin 66

Minimising nonlinear Raman crosstalk in future network overlays on legacy passive optical networks

D Piehler 70

Compact inductorless CMOS low-noise amplifier for reconfigurable radio

A Slimane F Haddad S Bourdel S Ahmed Tedjini-Baıliche M Tahar Belaroussi T Mohamed and H Barthelemy 74

Contents VOLUME 6 2014

The Best of IET and IBC Vol 6 2014

An electronic version of this publication and previous volumes can be found at wwwtheietorgibc or

by scanning the QR code

ISSN 2041-5923Introduction

Welcome to The Best of IET and IBC 2014 This is the sixth volume of an annual joint publication between the InternationalBroadcasting Convention and the Institution of Engineering and Technology

The IET is a formal member of the IBCrsquos partnership board but beyond this it has a long-standing and close relationshipwith the organisation through which they together encourage and promote professional excellence in the field of mediatechnology Nowhere is this relationship more strongly reflected than in the pages of this publication which celebrates thevery best technical media papers from this yearrsquos IBC Proceedings and the IETrsquos flagship journal Electronics Letters

This year our editorial takes a look at IBCrsquos Future Zone ndash one of the most exciting areas of exhibition space where theworldrsquos most hi-tech media companies and research organisations proudly demonstrate some of their very latest concepts andexperimental technologies Here you can not only see tomorrowrsquos media but you have the opportunity to experience itpersonally leaving impressions that will remain with you long after IBC

We then present eight papers chosen as the best contributions to IBC2014 by the IBC Technical Papers Committee and theexecutive team of the IET Multimedia Communications Network These include the overall winner of IBCrsquos award for the BestConference Paper lsquoDVB-T2 Lite Exploiting Broadcast HDTV Networks for Services to Mobilesrsquo and papers representing thehot topics of 2014 UHDTV (Ultra High Definition Television) (HDR) High Dynamic Range video HEVC (High EfficiencyVideo Coding) next generation studios and new directions in stop-motion film making

We are also pleased to present personal interviews both with an individual and with a research team whose significant workappears in this volume First Gino Alberico of RAI in Italy the principal author of IBC2014rsquos best paper and then with the teamat the Fraunhofer Institute for Integrated Circuits who have used advanced signal processing to inject novelty into the traditionalskill of stop-motion animation Find out the background on these individuals What motivates them What aspect of the workdid they enjoy most And if budgets were not a constraint where would they take the future of their work

From Electronics Letters this year we have three papers chosen from those published since IBC 2013

Electronics Letters has a very broad scope covering the whole range of electronics-related research and the three papers includedthis year deal both with research that may have a great impact on media technology in the long term as well as research aimed ataddressing the immediate issues of expanding services with existing infrastructure

The IBC papers you can read here represent the cream of almost 300 synopsis submissions originally received at our officesearly in the year ndash this is a figure that has remained remarkably constant for well over a decade We know that the writing andsubmission of papers for IBC is less often driven at the corporate level but usually represents the motivation of individualresearchers and teams who choose to write about their novel ideas experiments prototypes and trials and to share these inour conference forums We are extremely grateful and proud that so many media professionals continue to choose IBC forthe publication of their technical work and as a forum for discussion with their fellow workers This journal is a tribute to allthose individuals who submitted synopses this year whether successful or not If you are inspired by the papers and storiespresented here and would like to tell us about your own research then please look out for our call for papers in January eachyear And if your synopsis was unsuccessful this year then please try again ndash we work hard to accommodate as many papersand posters as we possibly can

1 The Best of IET and IBC 2014 Vol 6 pp 1ndash2amp The Institution of Engineering and Technology 2014

I hope that you enjoy reading this collection of the best papers as much as I and my committee of specialists and peer reviewersWe would like to convey our thanks to everyone involved in the creation of this yearrsquos volume both at the IET and at IBC and toextend our best wishes for a successful and exciting IBC 2014

Dr Nicolas LodgeChairman IBC Technical Papers Committee

Who we areIBC

IBC is committed to staging the worldrsquos best event for professionals involved in content creation managementand delivery for multimedia and entertainment services IBCrsquos key values are quality efficiency innovation andrespect for the industry it serves IBC brings the industry together in a professional and supportive environmentto learn discuss and promote current and future developments that are shaping the media world through ahighly respected peer-reviewed conference a comprehensive exhibition plus demonstrations of cutting edgeand disruptive technologies In particular the IBC conference offers delegates an exciting range of events and networkingopportunities to stimulate new business and momentum in our industry The IBC conference committee continues to craftan engaging programme in response to a strong message from the industry that this is an exciting period for revolutionarytechnologies and evolving business models

The IET

The IET is one of the worldrsquos leading professional societies for the engineering andtechnology community with more than 150000 members in 127 countries and offices inEurope North America and Asia-Pacific It is also a publisher whose portfolio includes asuite of 27 internationally renowned peer-reviewed journals covering the entire spectrumof electronic and electrical engineering and technology Many of the innovative productsthat find their way into the exhibition halls of IBC will have originated from research published in IET titles with morethan a third of the IETrsquos journals covering topics relevant to the IBC community (eg IET Image Processing ComputerVision Communications Information Security Microwave Antennas amp Propagation Optoelectronics Circuits amp Systems andSignal Processing) The IET Letters contained in this publication come from the IETrsquos flagship journal Electronics Letterswhich embraces all aspects of electronic engineering and technology Electronics Letters has a unique nature combining awide interdisciplinary readership with a short paper format and very rapid publication produced fortnightly in print andonline Many authors choose to publish their preliminary results in Electronics Letters even before presenting their results atconference because of the journalrsquos reputation for quality and speed In January 2010 Electronics Letters was given a freshnew look bringing its readers even more information about the research through a colour news section that includes authorinterviews and feature articles expanding on selected work from each issue

Working closely with the IET Journals team are the IET Communities team The communities exist to act as a natural homefor people who share a common interest in a topic area (regardless of geography) foster a community feeling of belonging andsupport dialogue between registrants the IET and each other Assisting each community is an executive team made up of willingvolunteers from that community who bring together their unique experience and expertise for the benefit of the group Membersof the Multimedia Communications Community executive team play an essential role in the creation of this publication inreviewing suggesting and helping to select content They contribute their industry perspectives and understanding to ensurea relevant and insightful publication for the broad community represented at IBC showing the key part volunteers have toplay in developing the reach and influence of the IET in its aim to share and advance knowledge throughout the globalscience engineering and technology community

The Best of IET and IBC 2014 Vol 6 pp 1ndash2 2

amp The Institution of Engineering and Technology 2014

ISSN 2041-5923

Editorial


The Future Zone has become the jewel in the crown of the IBC Show It is a fantastic collection of the latest cutting-edge projectsand prototypes from leading RampD labs around the World brought together into a single exhibition area in the Park Foyer next toHall 8 in the RAI Centre

In selecting exhibits for the Future Zone we try to select ideas and technologies from all sizes of companies ndash from large multi-national organisations to the smallest start-ups ndash and we start from the premise that the craziest proposals may well be the best Thereare only really four guidelines for the selection Committee

- the exhibit must be interesting innovative and valid

- it must not be a product that is already on the market ndash these can be shown on company stands in the other Exhibition Halls

- it should consist of a practical demonstration with hands-on interaction where possible

- its presentation should have a lsquowow-factorrsquo (measurable on the Richter Scale) to encourage passing visitors to stop and discovermore

This yearrsquos Future Zone exhibitors are truly international with representatives from a number of Japanese technology initiatives inboth broadcasting and communications MPEG and European collaborative projects Chinese broadcast science projects NorthAmerican industry developments and Korean media research amongst others There are incredible exhibits from some of themajor players at the forefront of our industry ndash like NHK BBC Cisco VRT

The practical demonstrations arraigned on the stands in the Future Zone this year include personalisation and enhancement ofthe viewer experience highlights of the FIFA World Cup in 8K UHDTV at 120 fps 4K UHDTV Cloud services future DigitalRadio broadcasting a day in our future lsquoConnected Lifersquo new technologies for AV archive restoration lsquosmartrsquo convergent media3608 viewable 3D displays and more A portal to view our future directions rather than just a glimpse of the future

The Future Zone is also home to IBC lsquoPostersrsquo These are the highly respected rigorously peer-reviewed ideas chosen by the IBCTechnical Papers Committee for their relevance to the subjects of IBCrsquos Conference Programme and their ability to startle yourimagination The Posters are displayed as two A0-size lsquosnap-shotsrsquo of innovation and this year we have created a great newviewing area with more of a modern lsquoart galleryrsquo presentation style This will enable visitors to view the displays moreconveniently and engage with the Poster authors in more open and attractive surroundings

We are especially pleased to have attracted Poster presentations for IBC2014 from so many different countries around the WorldThe fascinating topics of international research that will be presented by the authors from both industry and academia on the Fridayand Saturday Showdays include UHDTV-IP HEVC in the Studio Perception of 3D Audio CGI and editing for 3D images

3 The Best of IET and IBC 2014 Vol 6 pp 3ndash4


Crowd-sourced Music DTV Mobile Surveillance Object-oriented Storytelling Visual Quality Metrics 4K Broadcast on a BudgetAdaptive Bitrates amp Cloud-based Architectures for Multimedia and more

The Future Zone is open to all thoughout the IBC Show but we draw your attention in particular to the IET Reception held inthe Zone at 1600 on Friday the 12th September Here you will be able to go round the exhibits talk with the presenters andnetwork with the lsquomovers and shakersrsquo in our industry with some complimentary refreshments from the IET to help nourishyour stimulated mind

Professor David CrawfordIBC Conference Executive Producer




ISSN 2041-5923

DVB-T2 LITE exploiting broadcast HDTVnetworks for services to mobile receiversG Alberico A Bertella S Ripamonti M TaboneRAI - Research amp Technology Innovation Centre Italy

Abstract In several countries DVB-T2 networks are being deployed for the delivery of domestic television overterrestrial channels However there is also an increasing demand for linear TV viewing on mobile and portabledevices such as tablets Whereas 4G mobile networks could provide services to these terminals using unicastdelivery the cheapest delivery mode to big audiences during peak hours or during major events isundoubtedly via broadcasting For this reason DVB introduced a profile of its T2 standard called T2-Liteorientated towards mobile reception

To explore this further RAI launched an experimental trial in the Aosta Valley where the same transmittingnetwork was to be used to provide both HDTV services to domestic receivers and mobile TV services toportable devices

This paper reports the positive results emerging from the trial which proved that services to mobile devices arefeasible at marginal costs

IntroductionFollowing analogue television switch-off in many countriesDVB-T and DVB-T2 terrestrial networks are beingdeployed or planned for the delivery of digital televisionservices to domestic receivers Most of these networks aredesigned to provide services to fixed receiving systems withroof-top antennas but there is also an increasing demandfor TV viewing on mobile and portable devices such astablets and in-car screens

Mobile TV services such as those based on the DVB-Hstandard were launched some years ago but had limitedsuccess or even failed for many reasons Typicalproblems were low transmission efficiency high networkcosts unclear business models and poor terminalperformance

Nowadays major changes have occurred in the domain ofreceiving devices with the success of tablets and smart-phones people have devices that are capable of displayingmobile or nomadic video in a form that is truly compellingIn addition users are becoming accustomed to VoD andlive content consumption on a great range of devices

thanks to 3G4G networks and Wi-Fi connections whichare widely available at home or around

Multiple cell handovers and network capacity howevermake live video streaming over a wireless network a pooruser experience Furthermore big audiences are seriouslychallenging those mobile operatorsrsquo which attempt to servecustomers with unicast delivery Whenever 3G4Gnetworks reach capacity limits and are unable to deliver anacceptable quality of service a possible solution is (apartfrom using more spectrum or increasing the number oftransmitters) to off-load traffic onto other networktechnologies for example Wi-Fi or broadcast

From the technical and economical point of view broadcastnetworks are certainly the best delivery mechanism forreaching big audiences during peak hours or during majorevents For this reason DVB introduced in 2011 a newprofile of its T2 standard called T2-Lite aiming to bettersupport portable and mobile reception

Using the so-called FEF (Future Extension Frame) featureof the DVB-T2 system T2-Lite services can be added to T2broadcasts at the small expense of some capacity reduction



The T2-Lite services can operate on-top of the existinglsquoT2-Basersquo HD services in previously deployed networks andcan thus obviate the need to build new dedicated mobilenetworks

After experimenting in their laboratory with the firstprototypes of DVB T2-Lite RAI launched an experimentaltrial in Aosta Valley in 2013 Here HDTV services forfixed reception and T2-Lite mobile TV services wouldcoexist on the same frequency and would optimally exploitUHF spectrum while guaranteeing adequate signalrobustness for many potential additional services havingvery different requirements

The trial would validate the technical characteristics of thesystem an evaluation of mobile coverage a check oninteroperability related to the implementation of a SingleFrequency Network (SFN) using equipment from differentvendors and last but not least a test of the behaviour ofboth static and mobile reception in the field

The AOSTA VALLEY trialThe Aosta Valley region has very often been the scene ofexperiments by the RAI Research Centre especially sincethe advent of digital technology in the rsquo90s For examplein 1995-96 at the dawn of Digital Radio an intensivemeasurement campaign was carried out on DAB Whendigital terrestrial television was in its infancy in the latersquo90s the possibility of transferring the DVB-T signalthrough an analogue radio link was explored And again inthe early 2000s the Aosta Valley was the scene of animportant pre-operational test of DVB-T

The reason for using the Valley lies in its peculiartopography which is particularly complex from theviewpoint of signal propagation and reception Thenumerous side valleys and the large number of transmittersites needed to cover the area also make the Aosta Valleyperfect for testing single frequency networks Furthermorethe possibility of travelling along a dense secondary roadnetwork next to the main valley (which often climbs upsteep mountain slopes frequently hidden from the

transmitters) provides a demanding test case for the mobilereception of signals such as DAB digital radio or mobile TV

System architecture

The lsquobouquetrsquo of services for the trial originated in the head-end located at RAIrsquos Regional Centre in Aosta and consistedof three HD programmes for the T2-Base and of two servicesorientated to mobile reception for the T2-Lite The systemdiagram is shown in Figure 1

For the T2-Base a Transport Stream (TS) at about 287Mbits with three AVC-encoded live HD programmes wascreated using statistical multiplexing

For the T2-Lite a TS containing two AVC-encoded SDprogrammes was generated with a bit-rate between 15 and30 Mbits depending on the modulation scheme usedduring the tests

The two TS streams were sent to a flexible T2 Gatewaythat allowed the modulation scheme and the networkparameters for both the T2-Base and T2-Lite (Table 1) tobe chosen

For the T2-Base it was decided to use the 256QAMconstellation with FEC 34 allowing high transmissioncapacity while requiring a CN ratio of about 20 dB avalue similar to that required to receive current DVB-Tservices Since the T2-Base part is devoted to fixed

Figure 1 The head-end

Table 1 Modulation scheme for T2-Base and T2-Lite

Parameter DVB-T2 Base DVB-T2 Lite

Constellation 256QAM rotated QPSK rotated

FEC 34 13 ndash 12 ndash 23

FFT 32k 8k

Tg 1128 116

Guard interval 28 ms 56 ms

Pilot pattern PP7 PP4

Bit rate (Mbits) 287 15ndash20ndash30



reception with directional antennas the PP7 profile ofcarriers was selected The 32k FFT mode was chosen inorder to maximize the guard interval duration which isdifferent from that chosen for T2-Lite

For the T2-lite part a QPSK constellation was selectedand it was decided to perform the tests (in the first phaseat least) with three values of FEC 13 12 and 23 Formobile reception the optimal carrier profile found in thelaboratory tests was PP4 which provides excellent strengtheven at high reception speeds Finally the 8k mode wasused being a good compromise between robustness toDoppler shift and guard interval length

Finally the gateway generates the T2-MI streamcontaining all the information necessary for the set-up ofthe modulators placed at the transmitting sites With theaid of the time references 1 pps and a 10 MHz frequencyobtained by GPS all the synchronization information wasavailable for the implementation of a single frequencynetwork

Network set-up

Four transmitters (Gerdaz Salirod Arpy Courtil) wereinstalled in the Aosta Valley region and their locations areshown in Figure 2 Each transmitter was equipped with a

50 W power amplifier except for Salirod which has a100 W capability Horizontal polarization even though notideal for vehicle reception was used to take advantage ofradiations from existing antennas

The coverage simulations of the four transmitters allowedus to identify the areas of overlap of the various signals Astatic delay of 44 ms was applied to the transmitter inGerdaz of the T2-Lite stream in order to avoid inter-symbol interference In this way the zone of potentialinterference is moved into areas where signals are neversuperimposed In potentially lsquoat riskrsquo areas the interferencesare within the chosen guard interval of 56 ms

Laboratory testsVery extensive lab sessions allowed us to gain firsthandexperience of the enormous potential of the system and tounderstand how mature the technology is for promptexploitation

Tests were mainly conducted on a prototype boardequipped with one of the first available commercial chipsetscapable of receiving both T2-Base and T2-Lite

The tests employed all means for characterising thelsquogoodnessrsquo of the technology and its implementation onsilicon the sensitivity of the receiver the performance withwhite Gaussian noise (AWGN) the performance inRayleigh and Rice channel profiles to simulate respectivelyportable and fixed reception the performance in 0 dB echoto evaluate the resilience in reflections and SFNs the

Figure 3 Laboratory set-up

Figure 2 Geographical context

Figure 4 T2-Lite performances



effectiveness of the rotated constellations and theperformance with different mobile channels at differentspeeds (profiles COST206 Typical Urban Hilly Terrainand Rural Area) A full report of the results is available in(2) only in the Italian language

The laboratory set-up is depicted in Figure 3

In this paper only the data of interest for subsequenttesting in the service area are reported

The performance of T2-Lite in terms of CN at thethreshold of visibility (limited to the most robustmodulations suitable for mobile usage) are shown inFigure 4 The three FECs shown in the graph (13 12and 23) are also those used during the field tests Thevalues obtained are perfectly in line with what is expectedfrom the theoretical point of view with very smallimplementation margins

During the laboratory tests it was also observed that thetime-to-re-sync of the receiver is very short This is aparticularly appreciable feature in mobile reception wherenatural and man-made obstacles (bridges and tunnels forexample) can lead to a sudden loss of signal

Other laboratory tests that simulated the mobile channelaimed to verify the maximum speed that the system wasable to withstand

The graph shown in Figure 5 reports the required CNversus speed in kmh for UHF channel 53 (f frac14 730 MHz)which is the frequency used for testing in the Aosta Valley

From this graph it can be seen that the maximum allowedspeed with FEC 12 and 13 is around 240-280 kmh on achannel such as 53 which is located in the upper part of theUHF band At lower frequencies the maximum speed couldbe even higher compatible with a use not only in-car but alsoon high-speed trains

To obtain this result the pp4 profile of carriers was usedwhich compared to PP7 (commonly used for fixedreception) allows us to quadruple the maximum speed atthe price of 10 reduction of available bit-rate Theseresults are obtained with the 8k FFT mode If furtherincreases of maximum speed are required the 2k or 4kFFT mode could also be considered

The lab results confirmed the effectiveness of the new T2-Lite profile its efficacy on the mobile channel and the verygood performance of the first hardware implementations

Field testsThe tests in the service area were carried out by a car speciallyequipped for mobile measurements and by a van equippedwith a telescopic mast up to 15 m above ground level formeasurements at fixed locations

The purpose of the mobile measurements was to evaluatethe coverage area and to plot the map of field strength atthe roof of the car This data would be vital for futurenetwork planning

The purpose of the fixed-point measurements was todetermine the available field strength across the service areain order to verify consistency with the predictions obtainedthrough computer simulations The purpose was also to

Figure 5 T2-Lite performances on mobile channel



check whether the transmission is as expected with noproblems in the radiating equipment These measurementshave also allowed us to assess the margin of T2-Base signalreception In addition it was also possible to check all theSFN synchronization issues of the network verifying theparameters very accurately

Mobile measurements

In the first phase of the field tests the coverage of each singleT2-Lite transmitter was verified using three different valuesof FEC 13 12 and 23 For each FEC serviceavailability over more than 400 km of road was recordedthanks to the reception parameters logged by the T2-Liteevaluation board With these data recorded approximatelyonce per second and correlated with the GPS position ofthe car it was possible to get very detailed maps

Figure 6 shows the mobile reception availability with FEC12 for the Gerdaz transmitter Similar maps are available forall the other transmitters

The tests confirmed the excellent performance of the T2-Lite system that we had already seen in the laboratory Thedifferences in service availability due to FEC change arerather small between 13 and 12 and are slightly moreaccentuated with FEC 23

Overall despite the relatively low transmitting power(50 W) very good coverage was found on all types of road(motorway main and secondary roads) Also in urban areas(some densely populated) and in some tunnels results wereexcellent with complete absence of receiver unlocks orvideo artifacts On the highway no problems wereencountered up to the maximum permitted speed of130 kmh In this regard a test switching from PP4 toPP7 carrier profile was also performed and confirming thelaboratory results the receiver lost the signal when travelingat speeds higher than 70-80 kmh in a radial directionwith respect to the transmitter Supported by these findingsit can be reasonably assumed that T2-Lite should toleratespeeds up to 250 kmh thus opening interesting prospectsfor use on high-speed trains

During the data processing a statistical figure wascalculated which represents the value of EMF beyondwhich 95 of measurements fall on segments of 300 mThese values were plotted on the map with colours thatrecall the mobile reception availability thereby showing agraphical relationship between the electromagnetic fieldvalue and the service availability This information iscrucial during the network planning phase in order toproperly design the transmitter power and the radiatingsystem for the desired coverage To achieve this resultthe input power of a professional test receiver wasmeasured and given the antenna gain the available fieldstrength at the roof of the car could be calculated Alsoin this case the data acquired have been associated withthe location and speed of the car as obtained via GPSand the odometer

Finally the four-transmitter SFN was activated and afterthe necessary set-up tests the evaluation of mobile serviceavailability began Given the minimal differences incoverage obtained with FEC 13 and 12 values it wasdecided to use the latter one only being a bettercompromise between robustness and transmission capacityThe same routes used in previous tests were repeated thusgiving us a map of mobile service availability for the entirenetwork (Figure 7)

Fixed-point measurements

The fixed measurements carried out with the van equipped asdescribed above allowed us to detect the level of theelectromagnetic field available within the service areaThese data allowed us to check the reliability of thepredictions obtained with the computer simulation tocheck the evaluation of the reception margin of the T2-Base and to verify SFN synchronization

As was done with mobile measurements tests in the firstphase were carried out by activating one transmitter at a time

For each selected measurement point the electromagneticfield at different altitudes was measured in order to find themaximum value In addition so that we could have preciseinformation about the received signal and any anomaliesvarious reception parameters (such as MER constellationimpulse response etc) were also measured and recorded forall points

The measured field strength values are in very goodagreement with the computer simulations thus confirmingproper installation and operation of the transmittingequipment In general the electromagnetic field values arevery high providing a very wide reception margin across thewhole service area

Successively the four transmitters were switched on Firstof all the synchronization of the network was verified usingthe impulse response measured at several locations

Figure 6 Mobile reception availabilityFEC frac14 12 transmitter Gerdaz



Software developed by RaiWay was used to determine therelative delay between the two signals This software

operates from a knowledge of the coordinates of themeasuring point and any static delays present in the

Figure 7 Mobile reception availability for the full SFN (4 transmitters) FEC frac14 12

Figure 8 Example of fixed-point measurements



transmitters The comparison between the calculations andthe value obtained by the impulse response allowed us tovalidate the network

Then rotating the antenna in multiple positions for eachof the measuring points the spectrum of the signal theimpulse response an overview of the reception parameters(MER modulation scheme CN input power etc) andthe constellation were recorded for both T2-Base and T2-Lite

An example of the data collected during the measurementscampaign is shown in Figure 8

ConclusionsThe mixed T2-BaseT2-Lite field trials in Aosta Valleyconducted by the RAI Research Centre with the supportand collaboration of RaiWay led to the construction of ahigh-impact demonstrator for mobile TV

The field trials allowed the robustness of the T2-Litesystem to be evaluated in a region characterized by complexterrain where natural and man-made barriers oftenobstruct the view of the transmitters The sameconsiderations also applied to the tests carried out indensely populated urban areas like Aosta and Saint Vincentand in the several tunnels distributed along the routes

Using the so-called FEF (Future Extension Frame) featureof the DVB-T2 standard T2-Lite services and ldquoT2-BaserdquoHD services can coexist with previously deployed networksobviating the need for construction of new dedicatedmobile networks Potential infrastructure costs are thereforesignificantly reduced

At present the tests have been performed with a prototypereceiver and a car roof-top antenna so when targetingreceiving devices such as tablets with integrated antennashigher field strengths or a few additional sites might berequired to obtain a similar coverage

Summarizing T2-Lite showed during tests in thelaboratory and especially in the field its enormous strengthan awesome ability to receive signals under the most severeconditions

During the trial a huge amount of data on electromagneticfield strengths at car roof-top level vs mobile serviceavailability was collected representing an importantdatabase for future network planning of mobile services

Finally a serendipitous result which rewarded all the effortput in the initiative is that the test-bed is now being used bymany TV and chipset manufacturers to test their own T2-Base and T2-Lite receivers

AcknowledgmentsThe authors would like to thank their colleagues ArturoGallo Vittoria Mignone Bruno Sacco for the supportbefore and during the field tests and the colleagues ofRaiWay for the support in the main transmittersinstallation and for the field measurements Furthermorethe authors would also like to thank Screen ServiceBroadcast Technology SpA and Syes for the supportduring the installation of the transmitters

References

[1] ETSI EN 302 755 V131 Digital Video Broadcasting(DVB) Frame structure channel coding and modulationfor a second generation digital terrestrial televisionbroadcasting system (DVB-T2) EuropeanTelecommunications Standards Institute 2012

[2] A Bertella A Gallo S Ripamonti M Tabone 2013DVB-T2 e DVB-T2 Lite la sperimentazione in ValledrsquoAosta Elettronica e Telecomunicazioni Dicembre 2013pp 12ndash34



ISSN 2041-5923


1 Tell us a bit about yourself andwhat you do

I graduated in Elec-tronic Engineeringfrom Turin Polytech-nic and in 1988 Ijoined the RampDCentre of RAI Radio-televisione Italiana theItalian Public broad-caster where I amnow responsible forthe ldquoStudy andResearchrdquo department

I have beeninvolved togetherwith my research

team in the launch of digital TV and Radio services inItaly both on terrestrial and satellite platforms in thedeployment of multimedia services and interactiveapplications for connected TVs and second screen devicesalso combining TV and social networks

I am a member of the EBU Technical Committee andChairman of the EBU Strategic Programme on BroadcastInternet Services In my career I have also contributed tothe work of international standardisation bodies andindustrial forums such as ETSI and DVB as well asnational organisations like HD Forum Italy and Tivusatthe Italian free-sat platform

2 What is your paper aboutThe paper is about an experimental trial we have carried outin Aosta Valley showing that DVB-T2 HDTV services forfixed reception and T2-Lite mobile TV services forportable devices can coexist on the same frequency

Mobile services based on the T2-lite profile defined byDVB in 2011 are added on-top of existing ldquoT2-Baserdquo HDservices using the so-called FEF (Future Extension Frame)

feature of DVB-T2 At the expense of some capacityreduction this approach allows to exploit already deployedtransmitting networks avoiding to build a newinfrastructure dedicated exclusively to mobile services Thepositive results of the trial have been encouraging and haveconfirmed that providing broadcast services to mobiledevices is feasible at marginal costs

3 What do you find excitingabout this area of workThe most exciting aspect of this kind of activity is certainlytaking the first steps in the development of a newtechnology and finding the right technical solutions toproblems never addressed before Another amazing aspectof the trial is that our test-bed has been (and will be)used by different manufacturers for testing their T2-Baseand T2-Lite receivers Finally we are very satisfied of theresults obtained with T2-Lite technology the ability toreceive the signal in the most severe conditions is reallyawesome

4 What will you be hoping to seeat IBC this yearOf course I look forward to find tablets or portable devicesequipped with T2-lite front-end or at least with externalUSB adapters I am also very interested to see the latestdevelopments about HEVC video coding as well as cloud-based video encoding and streaming systems Anotherinteresting topic is about new immersive audiotechnologies spatial audio objects and their implications forthe home user

5 Do you believe that the futureof television will be mobile If sowill we soon see T2-Lite products inthe shopsI am convinced that digital terrestrial broadcasting is the bestdistribution platform for delivering high quality linear



services to large audiences and ensuring universal and free-to-air access for users

Itrsquos probably true that TV consumption will encompassmany different situations from the big screen in the livingroom to the nomadic viewing from live to on-demand somobile TV will be just one of the possibilitiescomplementary to one another Consequently there will becertainly a potential market for portable devices with a T2-lite capable frontend

6 Your paper describes both labwork and field trials Tell us whichyou found more challengingDifferently from the laboratory environment whereeverything can be kept under control field trial activity isdefinitely most challenging because of many additionalcomplications and challenges Just to mention a few ofthem synchronising the SFN network using equipmentfrom different vendors checking synchronisation in theservice area without at least in the first phase of the trialsuitable instruments adjusting transmitters relative delay inorder to avoid interference in overlapping areas and walking in high mountains in the snow to reach thetransmitting site for the installation of the equipment tofix a problem or to carry out a SW upgrade not feasibleremotely However we were lucky to work in a beautiful

region and this has allowed us to make lighter the obstaclesthat we have found during the trial

7 Given the high reception speedof T2-Lite will your future workinvolve collaboration with an Italiansupercar companyWe would be delighted to broaden our trial campaignincluding tests on a supercar but extending signal coverageto the Monza autodrome would be necessary to avoid finesfor exceeding speed limits on normal roads More seriouslywatching TV on a car running at 300 kmh may not berecommended unless yoursquore the passenger however similarreception conditions could be relevant for passengers onhigh speed trains

8 Your trials were conducted in abeautiful valley With an unlimitedbudget where else in the worldwould you like to travel to dofurther trialsUndoubtedly it would be fascinating to exploit radio wavespropagation and ldquocanyoningrdquo effects in Arizona perhaps bytesting mobile TV reception on a small plane glidingbetween the Grand Canyonrsquos walls




ISSN 2041-5923

Computational imaging for stop-motionanimated video productionsFrederik Zilly Matthias Ziegler Joachim KeinertMichael Schoberl Siegfried FoesselFraunhofer Institute for Integrated Circuits IIS Department Moving Pictures Technologies Germany

Abstract Creating movies using stop-motion technology remains a fascinating method of storytelling even in theera of digital cinema The huge number of professional and semi-professional clips produced as fan-art which canbe watched on the internet serve as testimony to the great interest in unleashing the creative potential of thisart-form However producing content using stop-motion remains a cumbersome task with high production costsespecially for full-length movies Consequently a trend can be observed that even many successful televisionseries which were originally produced using stop-motion have been moved to computer animation

Against this background we propose in this paper a new production scheme for stop-motion-animated movieswhich has the potential to lower production costs while increasing the quality of the resulting content By using astatic multi-camera array and algorithms from the field of computational imaging our technology permits thecreation of artistic effects in post-production which are difficult to realise using conventional stop-motionmethods Our approach allows changes in the depth-of-field smooth camera moves along virtual camerapaths and up-sampling of the frame-rate of the stop-motion-video all in high quality All effects arecomputed and applied in post-production while all intrinsic and extrinsic parameters of the cameras remainfixed during the whole production To demonstrate the practicability we shall show within this paper resultsfrom a stop-motion video which has been produced using the proposed methods

IntroductionHenry Selickrsquos famous stop-motion film Coraline publishedin 2009 is one of the most successful films shot with thistechnique It had a budget of about $60 million but wasable make about $125 million at the box office It is ofcourse not the only case of a successful stop-motion filmOther famous examples are productions of Wallace andGromit and Nightmare before Christmas Although shootingin stop-motion seems a little old-fashioned it is regularlyused for feature films as well as commercials and its slightstuttering appearance creates a unique impression on theaudience

Animation software suites constantly undergoimprovements however They are hard competitors when itcomes to the decision of whether a story shall be producedwith classical stop-motion techniques or using computer-generated stop-motion

In our opinion there are good arguments for bothapproaches Computer-generated stop- motion allowscomplete control to be exercised over the camera and sceneat all times Parameters can be changed and the finaloutput can be adjusted to suit the display platform (egcinema TV mobile platforms) On the other hand realstop-motion allows enthusiastic artists to use their fullcreative potential and the final result can score highly withits unique incomparable look

Classical stop-motion however also has some drawbackscompared with computer-generated animations The effortrequired to set up a scene and to shoot a take is veryhigh Moreover camera parameters must be fixed andcreating dolly shots camera pans or changing focus aretremendous tasks Therefore stop-motion tends to bemore costly and takes more production time comparedwith full animation

The Best of IET and IBC 2014 Vol 6 pp 14ndash19 14amp The Institution of Engineering and Technology 2014

Against this background we present a solution that cancircumvent some of the drawbacks mentioned whilepreserving the unique look of classical stop-motion filmsIn particular we propose using a camera array to capturethe light field of a scene and to postpone decisionsconcerning the camera set-up to post-production Suchdecisions can include changes of focus aperture or dollyshots

The remainder of this paper is organised as follows firstwe describe the legacy stop-motion production process itschallenges and limitations Then we present our approachwhich is based on a camera array and aims to overcomesome of the limitations of the legacy workflow The visualeffects which we make possible are explained and shownwith example images The fourth section presents details ofour stop-motion test shoot made with our proposedworkflow followed by an outlook and concluding remarks

Legacy stop-motion productionworkflowShooting a high quality stop-motion film is a tremendoustask that requires a lot of material time technicians and ndashmost important - enthusiastic artists As with every featurefilm or TV show producing a stop-motion film comprisesdifferent steps which are depicted in Figure 1 After theplot is completely defined the scene set-up and camerasettings are selected The film is shot and finished in post-production

The script is one of the most important factors in themaking of a successful film In the case of a stop-motionfilm every movement of the actors and every camera panneeds detailed planning In contrast to common movie-making it is generally not possible to shoot one sceneseveral times to gain several different camera positions orslight variations from the plot Because of the large effortand slow progress involved in the capture processre-shooting a scene to correct any production errors will becostly and time-intensive

From a technical point of view many requirements need tobe met in order to avoid visible artefacts in the final film Firstof all shooting a stop-motion video requires an indoor studioto maintain constant lighting conditions because smallvariations of luminance will cause visible flickering inthe film

Typically a scene set-up for a stop-motion film is spatially-limited as the actors in the film are puppets having sizesranging from a few centimetres to about 30 centimetres forhigh quality characters having sharp detail with naturallooks and movements But this spatial limitation also limitsthe number of cameras that can be squeezed into the scene

Beyond the requirements for scene set-up and puppets thecameras and supporting tools also have a strong influence on

Figure 1 Typical workflow in a stop-motion production

Figure 2 The camera set-up in a stop-motion environmentrequires definition of camera paths definition of focusand aperture and additional tools like gantries or tripodsneed to be included These set-up decisions must bemadein advance and cannot be changed



the resulting image quality Some of these aspects arementioned in Figure 2 and will be outlined in thefollowing As with live action feature films high qualityimages require high quality cameras Unfortunately suchcameras are usually physically large DSLR cameras aretypically a good choice for stop-motion as they can deliverthe required image quality Additionally each camera mustbe mounted on a robust tripod or similar device in order toavoid distortions by slight and random camera moves Alsodue to the size of the camera bodies the inter-axialdistance between the camerasrsquo optical centres is too largefor shooting stereoscopic 3D

With a camera mounted on a tripod it is possible togenerate a camera pan but it is not possible to move thecamera through the scene Generating precise dolly shots orshooting the film in 3D requires additional mechanicaldevices like precise sliders Depending on theirfunctionality these devices can increase the time necessaryto take a single frame as different camera positions need tobe captured and the camera needs to be precisely moved todifferent positions This can be done automatically byelectronic drives or manually

Another important aspect of the camera concerns the pointof focus or depth of field (DOF) which is a widely usedcinematic tool used to influence the visual impressionchanges of focus and DOF within a scene are oftendesired In the case of stop-motion special care must betaken to generate a continuous-looking change of focusAgain this can be accomplished by special tools which maybe added to the camera to adjust its focus with an externaldrive Similar considerations apply to the camerarsquos apertureif its optics allow a seamless change However thecombination of all the visual effects that are available in anormal film are hard to combine in a stop-motion set-upAchieving a natural look when using camera movementsand focus changes requires a lot of hard work and a lotof time

All the technical aspects of lighting scene and camera set-up which might hinder the artist from focussing on theartistic content must be considered In order to reducethese dependencies and simplify the shooting of stop-motion films we propose a new workflow that can enhancethe creative process

Proposed workflowInstead of using one or more DSLR cameras to shoot thescene we propose using a two- dimensional camera arrayfor image acquisition By using a camera array such as thatpresented by Foessel [1] and Zilly [2] we can capture thelight field of a scene and in combination with computervision algorithms we can simulate virtual cameras withinthe dimensions of the array Changes of the DOF can bepostponed to post-production and the artist doesnrsquot need toworry about the camera set-up when capturing the scene

Beside a camera array other implementations of light fieldcameras eg as presented by Ng [5] provide similarfunctionality but suffer from a lower baseline Thereforeflexibility to change the viewpoint of the virtual camera islimited compared with an array

An example of a two-dimensional camera array is shown inFigure 3 It comprises a set of small identical cameras that arearranged on a grid with identical line-of-sight and constantseparation distance This distance needs in general to beselected according to the scene In the case of a stop-motion scene with small objects the camera distance ischosen to be in the range of 4ndash6 cm in order to achievehigh image quality and minimize the probability of visibleartefacts When shooting the take the artist need not beconcerned about camera settings After each small changein the scene every camera in the array takes an imagetypically with a high DOF and therefore with a wide rangeof objects in sharp focus In our array every camera uses thesame settings such as focal point and exposure time

In order to understand the possibilities that arise from acamera array one needs to take a short look at its workingprinciple Due to their spatial distance every camera seesthe scene from a slightly different point of view Thesedifferences are used to employ the algorithms for depth-image-based rendering (DIBR) More details on light fieldprocessing can be found in [3] and [4]

The processing chain for the array is depicted in Figure 4After image acquisition image rectification is the firstcomputer vision algorithm to be used This corrects smallmisalignments of the cameras Precise alignment isimportant as it is required for the subsequent disparity

Figure 3 Camera array in a 4x4 set-up



estimation This estimation algorithm computes disparityfrom stereo image pairs in the vertical as well as thehorizontal direction These horizontal and vertical disparitymaps are then merged to remove occlusions and othernegative effects that can remain in the stereo disparity maps

This step results in a set of dense disparity maps for everycamera They contain an implicit 3D model of the capturedscene These two steps are the most basic ones and need to beexecuted for every image in a take All following processingsteps rely on this data and therefore these steps requirehigh precision Errors here can cause visible artefacts insubsequent processes

Based on disparity maps and rectified images a renderingalgorithm can compute parallax-compensated views fromarbitrary positions in the array The camera position can bedefined independently from a specific point in time as itwould be with a normal camera This enables us to movethe camera through the scene as a normal dolly would doAn example of a virtual dolly shot is depicted in Figure 5

In a typical dolly shot time moves forward linearly Withlight field processing however we can slow down or evenstop time and create a bullet-time effect that is well knownfrom the movie Matrix and other feature films

As mentioned earlier the array captures the scene withhigh DOF Using DIBR enables us to set the focus on aspecific object and also to change it with time This effectof course only works if DOF is reduced when comparedwith the original image This is done by adding a syntheticaperture to the virtual camera At the beginning of a scenewe can set the focus on an object and smoothly change itto another object as time continues Such a change of focusis depicted in Figure 6 At the beginning the backgroundis in focus (left image) changes to medium distance (centreimage) and is finally in the front in the right image

The dolly-zoom also known as Vertigo-effect is anothervisual effect that is well-known from Hitchcockrsquos VertigoIn this effect the camera moves backward while the focallength is changed such that the object in focus remains thesame size This is a challenging effect for a feature film andis even more challenging in a stop-motion film By usingour proposed workflow it is possible to move our virtualcamera forward and also backward Additionally we canselect the focal length of the virtual camera as required toachieve the desired effect in post-production An example

Figure 4 Lightfield processing chain Rectification anddisparity estimation need to be executed before virtualcamera positions can be rendered

Figure 5 Example of a virtual dolly shot The camera moves from a position in the left part of the scene to a viewpoint in theright part The shift is done in a parallax-compensated way Notice that the helicopter is seen from different perspectives

Figure 6 Example of a synthetic refocus Starting with focus on the background in the left image the focus moves to mediumdistance in the central image and ends on the foreground in the right image



image sequence with a Vertigo-effect is depicted in Figure 7It is especially good to see that the background seems to moveaway from the left to the right image while the policeman inthe foreground stays at the same size

Experimental stop-motion filmIn order to test our proposed workflow and also ouralgorithms in a real-world set-up we decided to make ashort stop-motion film Our target was to create a videowith a plot that covers about 20 seconds The story is abouta policeman drinking from a coffee cup in front of his

police station In the background a bad guy escapes fromhis prison sneaks to a helicopter and tries to get away At aframe-rate of 12fps this required at least 240 framesexcluding additional frames before and after each scene

The camera array employed has been seen in Figure 3 Itconsists of 16 identical cameras mounted in a 4x4 gridEach camera has a resolution of 2048 x 1536 pixels thatallows us to create the final film in full HD resolution Thedistance between cameras is about 6 cm A photo from thecomplete scene set-up is shown in Figure 8

The camera is at a distance of about 80 cm from the sceneFor the scene set-up we decided to use a mixture of LEGOand natural materials while the background consisted of aprinted poster

Shooting relevant raw data for the film took about 3 daysexcluding scene set-up and ended up with about 400 framesAs each frame consisted of 16 single images this gives 6400raw images As soon as the first take was available postprocessing could start using the proposed workflow and theimage processing algorithms we have described In post-production we created a film that contains all the visualeffects discussed and combines them with a simple narrative

During the development of all the necessary algorithms welearned that disparity estimation is the most critical point inthe overall light field processing chain This algorithm andsupporting filters need to work as precisely as possibleDisparity maps need to be dense with no missing disparity

Figure 7 Example images showing a Vertigo-effect The virtual camera moves backward while zooming in The policemanremains sharp and at same size while the background gets smaller

Figure 8 Making-of picture from the stop-motionproduction showing the scene in the background and thebackside of the 4x4 camera array in the foreground

Figure 9 Example image from a camera in the array and the corresponding disparity map Blue areas denote low disparitygreen denotes medium disparity and red areas are close to the camera



values but the values also need to be correct with littledifference from the ideal Any such differences can lead tovisible and disturbing artefacts in the final result Figure 9shows an example image as it is seen by one of the camerasin the array (left image) and the corresponding disparity map(right image) As can be seen the disparity map shows smallartefacts (eg on the ground and around the helicopter)

Nevertheless we were able successfully to test our proposedworkflow to design visual effects in post-production and tomodify them as necessary (Figures 5ndash7) Despite somedeficiencies in the disparity maps which can be reducedmanually if required the resulting image quality wasconvincing

ConclusionsStarting with a look at typical stop-motion production wehave described how shooting in stop-motion has somedrawbacks compared with computer-generated animationshowever the former remains popular due to its uniquecharacter Many of the difficulties discussed concernlimitations in the stop-motion camera system

By using a camera array we can overcome some of theselimitations and can even provide more creative opportunitiesfor artists both on the set and in post-production This isachieved by introducing techniques from computationalimaging which simplify the camera set-up and allow theartist to focus on the story We presented visual effects likedolly shots refocusing and synthetic aperture that can beapplied in post-production Additionally an example of thedolly-zoom has been presented further demonstrating thecreative possibilities available

However image quality is the most vital considerationwhen it comes to professional production and it was forthis reason that we tested our technology with a shortexperimental production The results proved the potentialof our workflow In our future developments we shallfurther improve our algorithms and acquisition systems tosimplify application of our tools and we shall also improveimage quality In the longer term we shall also adapt thistechnology for video acquisition to make it available to awider audience

References

[1] FOESSEL S ZILLY F SCHAFER P ZIEGLER M KEINERT J lsquoLight-field Acquisition and Processing System for FilmProductionsrsquo SMPTE 2013 Annual Technical Conference ampExhibition Los Angeles USA Oct 2013

[2] ZILLY F SCHOBERL M SCHAFER P ZIEGLER M KEINERT J FOESSEL SlsquoLightfield Media Production System using Sparse AngularSamplingrsquo ACM SIGGRAPH Annaheim USA 2013

[3] LEVOY M HANRAHAN P lsquoLight Field Renderingrsquo Proc ACMSIGGRAPH 1996 pp 31ndash42

[4] FEHN C lsquoA 3D-TV approach using depth-image-basedrendering (DIBR)rsquo Proceedings on Visualization Imagingand Image Processing 2003

[5] NG R LEVOY M BREDIF M DUVAL G HOROWITZ M HANRAHAN PlsquoLight Field Photography with a Hand-held PlenopticCamerarsquo Computer Science Technical Report StanfordUSA 2005



ISSN 2041-5923

Interview ndash Frauhofer Institute for IntegratedCircuits

1 Tell us a bit about your projectteam Is this the first time yoursquove allworked togetherThe group computational imaging amp algorithms was estab-lished 3 years ago It is part of the moving pictures technol-ogies department in the Fraunhofer institute for integratedcircuits IIS in Erlangen Germany Since then we performresearch in the field of high dynamic range imaging lightfieldimage processing and multi-camera image processing As ourfocus lies on applied research we are always looking foropportunities to create algorithms which enable new pro-duction workflows Our goal is to increase the creativenessof people in the movie business

2 What is your paper aboutWe have created algorithms which enable a multitude ofvisual effects to be applied in post-production Interestingexamples are refocusing virtual camera paths and thevertigo or dolly-zoom effect Thereby our algorithmsprocess data captured by a multi-camera rig In thiscontext we have captured several sequences of stop-motionusing a camera array consisting of 16 cameras Themultitude of perspectives can be used to generate newvirtual views Our paper describes the different effectswhich are possible when applying algorithms from the fieldof computational imaging

3 Why are you all so excitedabout stop-motion animationOn the one hand the technique of stop-motion animation issuitable to create a laboratory environment for multi-cameraimage processing developments The lighting conditions thesize of the scene and the position of the cameras and sceneobjects can be controlled precisely Moreover therequirements regarding hardware and equipment arerelatively low and good results can be achieved using

medium priced DLSR cameras On the other hand stop-motion animation has this special look Real objects withreal textures can be used eg natural stones are part of ourscenery which could not be generated in the same qualityusing computer animation While working with the stop-motion material we learned to love the unique combinationbetween reality and fiction which can be created usingstop-motion animation

4 Your technology brings stop-motion up to date but donrsquot youthink that it will eventually die outas an art-formBefore we started our project we assumed that only a verysmall number of stop-motion animated videos would beproduced today and that the remaining ones would becreated by fans and enthusiasts without any economicpressure We learned how vital this scene really is evengrowing due to the advent of medium priced DLSRcameras and prosumer solutions for PC-based video editingHowever stop-motion is also economically relevant for theadvertising industry and childrenrsquos programs Last but notleast the commercial success of the LEGO movie showthat there is a demand for this special look ndash even if themovie only looks like stop-motion animation while beingcomputer animation We believe that with newtechnologies such as our multi-camera image processingchain we can revitalize this art-form by adding newcreative options

5 Did making your experimentalproduction inspire any of you totake up stop-motion as a hobbyMany ideas came up in the recent past regarding interestingscene settings etc The hope is to be able to capture themusing a multi-camera array as we are excited about the newpossibilities which come with the multi-camera image



processing algorithms The inspirational part can be seen as ahobby and maybe some of us will start producing their ownstop-motion animations videos but this has not happenedyet

6 Are there any other traditionalstop-motion methods which mightbenefit from your technologyOur expectation is that the ability to change the focus andcamera position in post-production is helpful for manystop-motion techniques be it puppetoon brickfilm or eventime lapse movies We are looking forward to exploringnew fields of application and hopefully we will be able toapply our algorithms to material shot using additional stop-motion methods

7 You mention several classicstop-motion movies If you couldwork with any creative team in theworld which would it be and whatwould you hope to learnThere are so many teams and individuals all over the worldwhich spend a huge amount of personal effort and

enthusiasm and put all of their heart and soul into stop-motion animation projects so it is difficult to pick a singleteam However as we are doing applied research we wouldlike to learn more about specific requirements andworkflows as we try to understand what the artists needIn this context and because we also work in the field ofstereoscopic 3D production it would be interesting to workwith the team which created the movie Coraline in 3D ieLaika LCC It is our expectation that their approach tocapturing multiple images of the same scene using a singlecamera and a translation robot could be easily combinedwith our multi-camera image processing algorithms



Selected content from IBC 2014

ISSN 2041-5923

THE ART OF BETTER PIXELSDG BrooksDolby Laboratories Inc USA

Abstract This paper gives an overview of the key experiments used to determine the parameters required forbaseband video which combines both HDR (High Dynamic Range) and WCG (Wide Colour Gamut) We call thiscombination EDR (Extended Dynamic Range) for short

Furthermore it provides an insight into the EDR content creation process and proposes how the creative intentcan be propagated throughout the distribution chain all the way to the TV receiver in order to provide the viewerwith the best possible experience on a wide range of EDR-capable TV receivers

It is hard to exaggerate the impact that the cathode ray tube (CRT) had and still continues to have on thedesign of television standards Our CRT heritage affects our HDTV digital standards in a number of key ways(1) The colour gamut (as defined by the rare earth phosphors used) (2) The brightness was limited to 100candelam2 (units often referred to as ldquonitsrdquo) not only to control large area flicker but also to prevent theelectron beam from spreading and reducing spatial resolution

INTRODUCTIONThe quality of colour television has been dominated by thephysics of the cathode ray tube in determining the spatialresolution the image dynamic range the colour gamut andeven the frame rate of the images displayed

UltraHD (UHD) as specified in Recommendation ITU-RBT2020 was the first standard to partially break free fromthese constraints however it retained the 100 candelam2

CRT reference brightness limit and the correspondingElectro-Optical Transfer Function (EOTF) based on thegamma characteristic of the CRT

In April 2012 the United States made a submission toITU-R Working Party 6C which proposed a new EOTFin order to enable the carriage of high dynamic rangecontent through the 10 and 12-bit baseband interfacesspecified in the then draft version of ITU-R BT2020

The new perceptual EOTF proposed was based not uponthe gamma function of a CRT but directly on the contrastsensitivity ratio of the human eye as measured by Barten andreferenced in Report ITU-R BT2246-2 The submissionfurther proposed that based on viewer preference testing anHDR system should be capable of handling signals with abrightness range from 0 to 10000 candelam2

LOOKING TO THE FUTUREThere are 3 fundamental ways to improve moving imagequality

dagger Increase spatial resolution ndash more pixels

dagger Increase the frame rate ndash faster pixels

dagger Increase the capability of each pixel to represent morecolours and increased dynamic range ndash better pixels

The ITU-R UHDTV Recommendation BT2020 (1)published in August 2012 offered both more and fasterpixels with an increased colour gamut but did not considerthe need for pixels with an increased dynamic range andbrightness

This vestige of the CRT lives on despite the increasinginterest in high dynamic range systems in still photography

What dynamic range is required forentertainment content

Dolby developed the experimental display shown in Figure 1to answer this question This display consists of a digitalcinema projector pointed at a 23rdquo monochrome LCD



panel The digital projector image and the LCD panel aredual modulated to create a display capable of a black levelof 0004 nits and a peak white level of 20000 nits

Tests were performed to find out the preferred viewerexperience for

dagger Black level

dagger Diffuse White level

dagger Highlight level

Content was carefully chosen to ensure that the imagecontrast ratio did not change while testing with average

luminance level and largely achromatic images were used toavoid viewer preferences for more lsquocontrastyrsquo or colourfulimages

The full details of this experiment can be found in thepapers published by Daly et al (2) Daly et al (3) and DalyKunkel and Farrell (4)

From Figure 2 it is evident that the current TV standardsare inadequate as regards both dynamic range and brightnesswith consumer preferences being orders of magnitude greaterthan todayrsquos television systems can provide

Since the above results were first published there have beenquestions as to whether these diffuse white and highlight levelsare equally applicable to larger displays An additionalexperiment was performed with the same stimuli using a 4-metre screen and a digital cinema projector In bothexperiments the viewer was positioned 3 picture heightsfrom the screen The results of the later experiment arecompared with small screen results in Figure 3

Table 1 summarises these experiments and shows that adynamic range of 22 F-stops is needed to meet 90 ofviewer preferences for both small and large screenapplications One can further conclude that a televisionsystem which is capable of delivering content with adynamic range from 0001 nits to 10000 nits would satisfythe vast majority of viewers on a range of consumer devicesfrom tablets to very large LCD displays

COMBINING HIGH DYNAMIC RANGEWITH WIDER COLOUR GAMUTRecommendation ITU-R BT2020 (BT2020) providesnext-generation standard dynamic range TV systems with

Figure 1 Experimental HDR display

Figure 2 Summary of dynamic range subjective tests



a much wider colour gamut than current RecommendationITU-R BT709 (BT709) (5) consequently any newtelevision system must be capable of delivering both widercolour gamut and high dynamic range content

Colour Volume

Traditionally we have used the colour ldquohorseshoerdquo diagramto represent the colour gamut of a television signal suchas shown in Figure 4 however for each colour shownthere is a corresponding maximum luminance White isthe brightest colour as television uses an additive coloursystem

An alternative representation which allows both the colourgamut and the dynamic range to be represented is shown inFigure 5 Figure 5 also shows both todayrsquos BT709 colourspace and the new EDR colour space encompassing boththe high dynamic range requirement from 0 to 10000 nitsand the colour gamut as defined in BT2020

This representation has the advantage of showing allpossible colours at all available luminance levels

Having now defined the required colour volume needed tosatisfy the vast majority of viewers how can this berepresented in a practical television system

Figure 3 Summary of large screen dynamic range subjective tests

Table 1 Summary of Dynamic Range Experiments

Luminance level to satisfy 90 of Viewers

Large Screen Small Screen

Black 0002 nits 0005 nits

21 F-stops 20 F-stops

Diffuse White 3000 nits 4000 nits

1 F-stop 2 F-stops

Highlights 7000 nits 20000 nits

Figure 4 The colour ldquoHorseshoerdquo



EDR Colour Volume Quantization

What bit-depth is required to accurately represent this EDRcolour volume shown in Figure 5 To answer this questionrequires an investigation into both luminance andchrominance elements

Luminance Quantization

Report ITU-R BT2246-2 (6) makes use of Bartenrsquos modelfor the contrast sensitivity function of the human visualsystem to ensure that the contouring artefacts in theBT2020 UHDTV standard are acceptable

Using the existing Electro-Optical Transfer Function(EOTF) of Gamma 24 as defined in RecommendationITU-R BT1886 (7) would require a 15-bit representationto match the contrast sensitivity or contouring performanceof the human visual system as shown in Figure 6

However for a luminance range of 10000 nits Gamma24 does not accurately match the contrast sensitivityfunction of the human eye Gamma wastes bits by codingbright areas which are well below the Barten thresholdAlternatively using a 13-bit log representation wastes bitsin the darker regions

To resolve these issues a new Perceptual Quantizer (PQ)(8) was developed which follows the Barten curve to ensurethe optimum use of the bits available From Figure 6 itcan be seen that 12-bit PQ is below the visible thresholdfor contouring artefacts whereas 10-bit PQ is above

Colour Quantization

Whilst using PQ as the EOTF can provide the solution formonochrome images an additional metric is required toensure that colour images are also artefact free The JNDCross test Figure 7 was developed for this purpose

Each square in Figure 7 represents a block of pixels whichhave been perturbed by 1 code word away from the greybackground All possible variations of this in RGB result in26 different colour patches

If the picture looks like a flat grey image then there is novisible quantization If some patches are visible then somecontouring or banding is likely

This test is then performed for multiple different greylevels The industry-accepted metric to judge the visibilityof these patches is CIEDE2000 (9) colour differenceformula

Subjective tests by Nezamabadi et al (10) indicate that fornoise-free images such as animation CGI and graphics aDE2000 threshold of 25 is needed to avoid contouring

Figure 5 Colour volume EDR v BT709

Figure 6 Bartenrsquos model over 10000 nit range



artefacts When noise is present this threshold increases to5

It should also be noted that the DE2000 metric startsto lose accuracy below about 3 nits this corresponds to thepoint where the human visual system begins to switchbetween photopic and scotopic vision

Combining the use of this metric with the JND colourpatches allows the quantification of possible colourquantization artefacts The results are shown in Figure 9

EDR Colour Volume Conclusions

For noise-free EDR images such as animation CGI orgraphics using 12-bit PQ ensures that bothmonochromatic and colour banding artefacts are below the

visible threshold This matches the MovieLabs specificationfor next-generation content (11)

For captured images that contain some noise using a 10-bit PQ representation is adequate but care needs to be takenespecially with graphics However this is an area wellunderstood by broadcasters who deal daily with these issuesfor todayrsquos 8-bit BT709 transmissions

Based on the above studies a detailed submission wasmade in July 2012 by the United States to ITU-RWorking Party 6C proposing the PQ EOTF for HighDynamic RangeWider Colour Gamut images (12) ARapporteur Group was subsequently set up to examine theoptions Work in this group is on-going In parallel withthis effort the PQ EOTF is in the process of being

Figure 7 The JND cross (colours exaggerated)

Figure 8 CIEDE2000 overview



standardized by the Society of Motion Picture and TelevisionEngineers (SMPTE) (8)

GETTING EDR CONTENT TO THEHOMEHaving established the definition for EDR content how bestcan this content be transmitted to the home and presented tothe viewer

The following criteria were defined to meet thisrequirement

dagger Single production workflow required for both SDR andEDR content

dagger Compatibility with existing off-line and real-timeinfrastructures

dagger The highest quality images matching the creative intentshould always be available to the home viewer

dagger Image quality will increase in both professional andconsumer applications as display technology evolves - thetransmission system must not be the bottleneck

dagger The transmission system should allow backwardscompatibility with todayrsquos BT709 or BT2020 systems

dagger Should be independent of any spatial resolution colourgamut or frame rate

dagger The EDR transmission system should use industrystandard codecs

dagger A bit-rate-efficient transmission method is required forEDR

dagger New EDR TVs must be capable of mapping thetransmitted EDR images to the TV displayrsquos native colourvolume (defined by the displayrsquos black level peak whitelevel colour temperature and colour primaries) therebyremaining as true as possible to the original creative intentwithin the confines of the displayrsquos capabilities

An end-to-end system overview

Figure 10 provides an end-to-end architecture workflow foroff-line content creation and distribution which meets therequirements described above

The ability to create EDR content has been limited not bytodayrsquos cameras but by existing 100nit broadcast referencemonitors The development of the Pulsar monitor whichhas a peak brightness of 4000nits with a DCI-P3 colourgamut plus a set of EDR tools as plug-ins to commonlyused off-line production suites meet the needs of thecreative

In parallel with this process image metadata isautomatically generated and stored along with any creativeinput This metadata is used downstream to guide thedisplay management block in the TV receiver based onboth automatic and creative input captured duringproduction The definition of this metadata has begun andis now in the process of becoming a SMPTE standard (13)

To meet the practical requirements of backwardscompatibility with todayrsquos BT709 and future BT2020transmission systems a dual-layer codec architecture hasbeen developed which allows the existing standards to betransmitted as the base layer An enhancement layercontains the information necessary to recreate the EDRsignal Figure 11 shows a block diagram of thisarchitecture It is worth noting that this dual-layer codecworks outside of the coding loop (unlike MVC or SVC

Figure 9 EDR evaluation using DE2000



for example) so that no changes are required to eitherencoders or decoders

Moreover the dual layer codec as shown in Figure 11allows for either an 8-bit AVC or an 810-bit HEVC baselayer for compatibility with existing services Similarly theEDR input signal can be either 10-bit PQ or 12-bit PQand either will be faithfully reconstructed at the dualdecoder output

Using this dual-layer codec also allows existing TVs toreceive the base layer standard dynamic range signal whilstnew EDR televisions can receive the EDR signal and adaptthe content to match the capabilities of the display TheEDR enhancement layer only increases the bit-rate by25 compared with the backwards-compatible base layer

The capabilities of the EDR TV will evolve over time asnew technologies such as quantum dots allow wider colourprimaries and brighter displays By the use of acombination of televisionrsquos inherent display characteristicsplus the EDR metadata transmitted the displaymanagement function inside the TV is able to adaptoptimally the incoming EDR images to match eachdisplayrsquos characteristics whilst maintaining the creative intent

CONCLUSIONThis paper outlines the experiments made to determine theimage dynamic range required for television andcinematographic entertainment purposes

Using this information a series of tests and experimentswere performed to determine the baseband bit-depth

Figure 11 The EDR dual-layer codec

Figure 10 Off-line EDR system architectureworkflow


required to ensure that no contouring artefacts would bevisible

In making these tests the traditional TV gamma 24 non-linear curve was proven to be inefficient for high dynamicrange images and a new Perceptual Quantizer was designedto match the characteristics of the human visual system

Both 10 and 12-bit PQ EDR signals can be encoded bythe dual-layer codec proposed The SDR base layer can becoded as either 8 or 10-bit depending on the backwardscompatibility requirements

Existing SDR TV receivers will ignore the enhancementlayer and display the SDR content as today

New EDR receivers will decode both base andenhancement layers and combine these together tofaithfully reproduce EDR images at either 10 or 12-bit PQdepending on the source The TV receiver will alsoincorporate a new display management block which mapsthe content to the evolving characteristics of each displayby using the metadata present in the EDR stream

ACKNOWLEDGEMENTSThe author would like to thank his colleagues for all theirwork in developing EDR (Dolby Vision) from ImageResearch amp Development in Burbank SunnyvaleSan Francisco and Yardley I would also like to thank mywife Naomi for encouraging me to write this paper over abeautiful long spring weekend

Figure 4 is courtesy of Richard Salmon at the BBC

REFERENCES

[1] Recommendation ITU-R BT2020 lsquoParameter values forultra-high definition television systems for production andinternational programme exchangersquo August 2012

[2] DALY S KUNKEL T SUN X FARRELL S CRUM P lsquoPreferencelimits of the visual dynamic range for ultra high qualityand aesthetic conveyancersquo SPIE Electronic Imaging HVEIBurlingame CA Feb 2013 2013

[3] SCOTT DALY TIMO KUNKEL XING SUN SUZANNE FARRELL POPPY

CRUM lsquoViewer Preferences for Shadow Diffuse Specularand Emissive Luminance Limits of High Dynamic RangeDisplaysrsquo SID Display Week paper 411 VancouverCanada (2013)

[4] SCOTT DALY TIMO KUNKEL SUZANNE FARRELL lsquoUsing ImageDimension Dissection to Study Viewer Preferences ofLuminance Range in Displayed Imageryrsquo Invited talk IDWInternational Workshop on Displays SID Sapporo Japan6 Dec 2013 (2013)

[5] Recommendation ITU-R BT709-5 lsquoParameter valuesfor the HDTV standards for production and internationalprogramme exchangersquo April 2002

[6] Report ITU-R BT2246-2 lsquoThe present state of ultra-high definition televisionrsquo November 2012

[7] Recommendation ITU-R BT1886 lsquoReference electro-optical transfer function for flat panel displays used inHDTV studio productionrsquo March 2011

[8] Draft SMPTE Standard 2084 lsquoHigh Dynamic RangeElectro-Optical Transfer Function of Mastering ReferenceDisplaysrsquo

[9] CIE Publication 1422001 ldquoImprovements to industrialcolour difference evaluationrdquo Commission Internationalede lrsquoEclairage Vienna Austria (2001)

[10] NEZAMABADI M MILLER S DALY S ATKINS R lsquoColor signalencoding for high dynamic range and wide color gamutbased on human perceptionrsquo Color Imaging at SPIErsquo sElectronic Imaging Conference (2014)

[11] MovieLabs lsquoSpecification for Next Generation Videorsquo ndashVersion 10 2013

[12] Input document USA002 to ITU-R Working Party 6ClsquoParameter values for UHDTV systems for production andinternational programme exchangersquo April 2012

[13] Draft SMPTE Standard 2086 lsquoMastering Display ColorVolume Metadata Supporting High Luminance and WideColor Gamut Imagesrsquo




ISSN 2041-5923

Source-timed SDN video switchingof packetized videoThomas G EdwardsFOX Network Engineering amp Operations USA

Abstract The broadcast industry is considering a move from a serial digital interface (SDI) infrastructure to acomputer-networked infrastructure Professional media networking will enhance flexibility and agility of thebroadcast plant reduce and simplify cabling provide for format agnostic transport and will allow mediaorganizations to benefit from the economies of scale of COTS networking hardware

One key element of video processing is synchronous (clean) switching between video sources The accuracy ofthe SMPTE RP 168 recommended switch point is of the order of 10 ms Unfortunately the current speed ofnetworking switch flow-rule changes using software defined networking (SDN) is of the order of 1 ms to10 ms and the speed of rule changes is not highly deterministic

To solve this problem source-timed video switching separates the actions of updating SDN flow-rules with thetiming of the actual video switch This technique first adds new SDN flow-rules to the networking switch thatmatch a packet header value that is not currently being transmitted Then the video source changes itsoutputrsquos packet header value to the new value in the flow-rule at the precise RP 168 video switch point Thenew packet header value matches the new flow-rules thus the effect of these two steps achieves preciselytimed synchronous video switching (which can be compared with the actions of ldquopreviewrdquo and ldquotakerdquo on avideo switcher)

FOX has developed a proof-of-concept of source-timed video switching using live HD-SDI sources packetizedusing SMPTE 2022-6 and synchronously switched on a COTS 10 Gbps Ethernet switch

IntroductionProfessional media networking is the use by professionals of anetworked infrastructure to process media Although thebroad definition includes file-based workflows the greatestchallenge of professional media networking today is thecarriage of real-time audio and video flows over IT networks

Ethernet switches are readily available today at affordableprices that can switch terabits per second of traffic in anon-blocking fashion For example a 50-port 10 gigabitper second Ethernet (10 GbE) switch carrying six HD-SDI flows bi-directionally per port can serve as a300 times 300 uncompressed HD-SDI non-blocking videorouter

One of the key elements of video processing is thesynchronous (ldquocleanrdquo) switching between video sources If

video is switched in the middle of a frame there will beundesirable video artifacts thus it is desirable for the switchto occur during the vertical blanking interval (VBI) Thereis often ancillary data transmitted during the VBI so careneeds to be taken not to damage that data

SMPTE has defined [1] a switching point inRecommended Practice 168 (RP 168) such that the effectsof any signal discontinuity in the processing chain due toswitching is minimized The line designated for theswitching point is chosen to be after the vertical sync butearly in the VBI This is to ensure that ancillary signalstransmitted during the vertical interval remain with thevideo frame that they are associated with Also ancillarydata is excluded from the line following the switch line toallow for receivers to synchronize to EAVSAV (EndStartof Active Video) before data becomes present to avoid itsloss This exclusion is reflected in several SMPTE ancillary



data standards Finally the recommended switch area on theswitching line is a region relatively within the middle of theactive video to provide some tolerance for inaccurate timingand also to provide a reasonable ldquoguard areardquo to protect theSAV and EAV data flags

With packetized video it is unclear how many of therestrictions in RP 168 remain truly relevant As every bitshould survive transport within a clearly defined timingmatrix there are no concerns about synchronization toEAVSAV And as video packets should fully tile the rasterwithout bit-timing jitter the switch could occur earlier inthe VBI (if care is taken not to bisect ANC data at packetboundaries) However until our experience allows us toloosen the tolerances on the switching point it may bewise to implement switching in the RP 168 area

This paper will examine the different methods ofperforming synchronous switching of uncompressed videoand will concentrate on a novel method of switching thatutilizes a relatively new networking technique known assoftware defined networking (SDN) In SDN Ethernetswitches expose an API such as OpenFlow [2] to allow forprecise control of packet flow-rules including matchingpacket headers on layer 2 layer 3 and TCPUDP fieldsand then performing actions on the matched packets (suchas sending them to a specific output port dropping themre-writing a packet header etc) SDN breaks the ldquooldmodelrdquo of networking where switches implemented theirown specific algorithms for setting up packet-forwardingrules (such as spanning tree and IP routing protocols) and

replaces them by having a separate ldquocontrollerrdquo thatmonitors the network as a whole is directed by applications(such as broadcast-specific applications) and then sets upflow-rules on Ethernet switches in the data center toachieve the desired application-directed goals

SMPTE 2022-6SMPTE 2022-6 [3] carries the entire SDI signal includingall ancillary data (such as embedded audio ClosedCaptioning data and Active Format Descriptor) in anuncompressed fashion It uses an RTP payload to deliverthe SDI data using UDP The High Bit Rate Media(HBRM) Payload Header of 2022-6 includes video formatidentification a frame count a sample clock frequency anda video timestamp FOX has developed a Wireshark Luapacket dissector for 2022-6 available at httpsgithubcomFOXNEOAdvancedTechnologysmpte2022-6-dissectorAn example of the output of this dissector can be seen inFigure 1

The payload for each RTP datagram in 2022-6 is 1376octets of SDI data Since SDI typically has 10-bit samplesoften a sample is split between two RTP datagrams Thelast datagram of the frame ends with zero padding if thereis not enough SDI data left in the frame to fill all 1376octets Also the last RTP datagram in a frame has itsmarker bit set Total data overhead for 2022-6 carried byEthernet over the original SDI bit stream is about 54 to59 depending on header options Due to the singleuniform size of the datagram for all video formats there

Figure 1 SMPTE 2022-6 Wireshark dissection



typically is not an integer number of RTP datagrams per line(see Figure 2 the diagonal white lines are RTP datagramboundaries of 2022-6)

Switching 2022-6 in the RP 168 area on the boundary ofdatagrams can be achieved by counting datagrams from theend-of-frame set RTP marker bit as in Laabs [4] Table 1lists valid RTP datagram boundaries in the RP 168 area for2022-6 streams

Through the efforts of the Video Services Forum there hasbeen extensive work on interoperability between multiplevendors for 2022-6 streams as described in Kouadio [5]Although 2022-6 was originally conceived of as an inter-facility data stream some vendors are now exploring its usewithin the broadcast plant IP cores for conversionsbetween SDI and 2022-6 are commercially available for themajor FPGA types

Methods of switching packetizedvideoThree potential strategies for synchronous switching ofreal-time packetized video streams are shown in Figure 3

Previous work has attempted to solve the problem ofprecisely timed synchronous video switching throughprecise timing of packet flow-rule changes at the Ethernetswitch (ldquoswitch-timedrdquo) or by using buffering to allow forflow-rules to be precisely changed at the destination(ldquodestination-timedrdquo) Unfortunately both of thesetechniques have significant real world drawbacks

Switch-timed video switching

The switch-timed video switching technique suffers from thefact that today common-off-the-shelf (COTS) Ethernetswitches generally cannot provide the temporal accuracy offlow changes required for synchronous video switching Tochange flows within the RP 168 area for 720p5994 videothe required accuracy would need to be within the lengthof time represented by one 2022-6 datagram or about74 ms Unfortunately reported OpenFlow rule update ratesof COTS network devices from Bilbery et al [6] andMogul et al [7] range from 100 to 1000 per secondimplying an upward limit of temporal precision between1 ms and 10 ms It is possible that packet-forwarding ruleupdate rates on COTS network devices will become fasterover time but they are not fast enough for reliablesynchronous video switching today

Destination-timed video switching

Destination-timed video switching requires the destination toreceive the video flow to be switched into before ending thereception of the video flow to be switched out of (ldquomakebefore breakrdquo) During that time the destination buffers datafrom both video flows and the destination device itself candetermine the precise video switch point This solution canbe readily implemented using IGMP as in Laabs [4]However during the period between joining the newmulticast group and leaving the old multicast group one ormore data paths through the network will have to carrytwice the bandwidth of the media flow In particular theldquolast milerdquo Ethernet switch port that the receiving device isdirectly attached to is likely to be limited to handling halfof the media flows that it could otherwise carry in a

Figure 2 SMPTE 2022-6 datagram boundaries

Table 1 RP 168 switch points for 2022-6

720p 1080i (F1) 1080i (F2)

RTP Datagram rsquos in frame 20ndash21 26ndash27 2272ndash2273

Boundary pixels after SAV end 738 8314 6298



steady-state environment to avoid potential blockingrepresenting a significant waste of bandwidth capacity

Source-timed video switching

The concept of source-timed video switching is to separatethe temporally inaccurate process of updating Ethernetswitch packet-forwarding rules from the actual precisetiming of a synchronous video switch To do this a specificelement of the packet header is selected as the ldquotimingsignalrdquo match field whose value will be used to trigger aprecise flow change by matching rules previously configuredusing SDN on the network devices Preferably this headerwill have little or no impact on other stream processingfunctions In our proof-of-concept the UDP source portvalue was used

Figure 4 shows the initial state of an example source-timedvideo switching network There are two video flow sources

and two video flow destinations A notional ldquovideo routingtablerdquo (VRT) shows the relationship between flow sourcesand destinations The VRT has a version number (in thiscase 10001) which will be used to match the timing signalmatch field of the packet flow-rules Initially in thisnetwork a flow from Source A is sent to Destination Aand a flow from Source B is sent to Destination B Theflows have the UDP source port 10001 which matches theflow-rules on the switch to carry out the routing as directedby the VRT This example only shows one-to-one flowsbut the flows could also be one-to-many

A request to change video flows is received by the SDNcontroller in Figure 5 The notional VRT is first updatedto reflect flows from Source A to be sent to Destination Band flows from Source B to be sent to Destination A ThisVRT has the new version number 10002 The controllerresponds by adding new flow table rules to the Ethernetswitch to steer packets that have UDP source ports

Figure 3 Three ways to switch packetized video

Figure 4 Source-timed switching initial state



matching the new VRT version number (a process that couldtake tens of milliseconds) During this time the sources arestill transmitting flows with UDP source port 10001 sothe flows are still following the paths as per VRT version10001

After enough time has passed for the new flow-rules tohave been reliably added to the Ethernet switch usingSDN the controller informs the sources of the new VRTversion number as illustrated in Figure 6 This prepares thesources to change the UDP source port of their emittedflows at the precise time of the next RP 168 videoswitching point

In Figure 7 the sources have changed their UDPsource port to 10002 at the precise video switchingpoint and since the flows now match the new flow-rules

on the switch they are steered as directed by FRT version10002

After this point in time the old (now unused) flow-rulesmay be removed from the switch either programmaticallyor using OpenFlow idle_timeout

Proof of conceptA proof-of-concept was developed to test and demonstratesource-timed SDN synchronous video switching

Hardware

A schematic diagram of the proof-of-concept hardware isshown in Figure 8 Uncompressed HD-SDI video sourceswere from an SSD-based player and a live camera Both

Figure 5 Source-timed switching switch request

Figure 6 Source-timed switching VRT version update



video sources emitted 720p5994 fps video and they weresynchronized using a signal from a blackburst generator

The uncompressed HD-SDI video from the sources waspacketized using a 2022-6 gateway The gateway device hasfour HD-SDI inputs four HD-SDI outputs and two 10GbE SFP+ ports One gateway was used for packetizationof both HD-SDI inputs which allowed RTP and 2022-6HBRM metadata to be consistent between the twostreams This gateway also de-packetized one video flowoutput from a particular port of the Ethernet switch intoHD-SDI A second gateway was used only for a de-packetization of a second video flow output from anotherport of the Ethernet switch HD-SDI outputs from thegateways were displayed on professional monitors

A server was used to host software implementing thesource-timed packet header changes as well as SDN

orchestration The server had one 33 GHz Intel XeonE3-1230 V2 processor 8GB of RAM and ran UbuntuLinux Its Ethernet card was a dual Intel 82599EB-based10 Gbps SFP+ NIC The NIC was operated using ldquoDNA(Direct NIC Access)rdquo DMA-enabled drivers fromNtoporg Attached to the server was a USB-connectedbutton box that emulated pressing a space bar on akeyboard to command a switch of the video flows

The Ethernet switch was an Arista 7050S-52 a 52-port10 Gbps Ethernet switch with SFP+ interfaces and a1GbE out-of-band management (control) port

Server software

A controller was written in Python to orchestrate the source-timed flow change operation When the button box waspushed the controller contacted a DirectFlow agent on the

Figure 7 Source-timed switching sources change VRT version

Figure 8 Source-timed switching proof of concept schematic diagram



switch to install new flow-rules After a delay of 100 ms thecontroller then commanded a source-timed packet headeradjustor application (also running on the server) to changethe UDP source ports of the video flows at the next RP168 switch point After another 100 ms delay thecontroller then contacted the DirectFlow agent on theswitch to remove old flow-rules

The source-timed packet header adjustor receivedincoming 2022-6 video flows from the sources on one NICof the server and re-broadcasted them out of another NICwith modified UDP source ports The adjustor was writtenin C using the Ntoporg PF_RING high-speed packetprocessing socket library

Flow Steering with DirectFlow Agent

Aristarsquos DirectFlow API allows custom flow-rules to beadded by agent processes running on a switch usingsemantics similar to OpenFlow In this case DirectFlow-rules were used to match multicast UDP traffic from aspecific IP source address and UDP source port and todirect that flow to a specific destination Ethernet portwhile also changing the VLAN

A Python agent script was scheduled to run when theswitch is first powered The script listens on the 1GbEout-of-band management port for UDP command packetsThe payload of the UDP packet contains instructions fromthe orchestration system to the switch Upon detecting avalid new command packet the Python script adds orremoves the appropriate DirectFlow-rules via the AristaEOS Python API

Figure 9 shows an example of a flow-rule that directs UDPdatagrams with UDP source port 20001 from IP source10101052 to be output on port Ethernet 17 with VLANID 2

Results

The proof-of-concept delivered a system capable ofsynchronous switching of real-time uncompressed HDvideo sources at the RP 168 switch point using a COTSEthernet switch PCAP captures of the video switchingpoints were made at the output ports of the Ethernetswitch to show a clean switch from one source to anotherat the proper point in the video raster In casual testing

among users in our lab and at a broadcast industry tradeshow no ldquobad switchesrdquo were detected

In automated ldquotorture testingrdquo of switching as fast aspossible for many hours there were some occasional videoglitches in the switching of the order of one out of 10000switches This may be due to occasional de-synchronizationof the 2022-6 flows possibly leading to one too few or onetoo many datagrams for a frame at the destination The2022-6 standard contains no metadata for de-packetizers toknow where in the video raster a particular datagram comesfrom The only hint regarding a datagramrsquos location in theraster is that the datagram that ends a frame has its RTPmarker bit set Additional metadata in 2022-6 that informsde-packetizers about the position of datagrams in the rastercould allow them to better handle lost datagrams or thosethat duplicate data in a particular section of the raster

ConclusionsSource-timed SDN video switching is a way to achieve hightemporal accuracy of packetized flow changes despite thecomparatively slow and temporally inaccurate nature oftypical SDN APIs Also it does not require the ldquodoublebandwidth penaltyrdquo of destination-timed switchingPhilosophically it shows how there can be a beneficialcollaboration between COTS networking devices built forgeneral enterprises that can process billions of packets persecond according to slowly changing rules and the videodevices from our specialized industry that must be highlytemporally accurate

AcknowledgmentsThe author would like to thank Warren Belkin and AndreasBechtolsheim from Arista Networks for support inimplementing the PoC Ntoporg for several key elementsof software in the PoC Steven Wallace from IndianaUniversity InCNTRE for discussions about SDN andOpenFlow Al Kovalick for his suggestions regardingpacketized video switching theory and Clyde Smith andRichard Friedel from FOX Network Engineering ampOperations for providing the resources and guidance tocarry out this work

References

[1] SMPTE RP 1682009 Definition of Vertical IntervalSwitching Pont for Synchronous Video Switching

[2] Open Networking Foundation 2013 OpenFlow SwitchSpecification Version 133 httpswwwopennetworkingorgimagesstoriesdownloadssdn-resourcesonf-specificationsopenflowopenflow-spec-v133pdf

[3] SMPTE 2022-62012 Transport of High Bit Rate MediaSignals over IP Networks (HBRMT)Figure 9 Example DirectFlow-rule



[4] LAABS M lsquoSDI Over IP ndash Seamless Signal Switching inSMPTE 2022-6 and a Novel Multicast Routing ConceptrsquoEBU Technical Review 2012 Q4

[5] KOUADIO A lsquoSDI over IP interoperability testsrsquopresented at the EBU Network Technology Seminar 2013Geneva CH June 2013

[6] BILBERRY J PALMER M SULLIVAN R lsquoNetworkService Security Through Software-Defined Networkingrsquo

httpinstitutelanlgovistisummer-schoolcluster-student-projectsnetwork-service-security-through-software-defined-networking

[7] MOGUL JC TOURRILHES J YALAGANDULA P SHARMA P CURTIS

AR BANERJEE S lsquoDevoflow Cost-Effective FlowManagement for High Performance EnterpriseNetworksrsquo Proc 9th ACM SIGCOMM Workshopon Hot Topics in Networks Hotnets-IX 2010pp 11ndash16




ISSN 2041-5923

Live event experiences - interactive UHDTV onmobile devicesOA Niamut1 GA Thomas2 E Thomas1

R van Brandenburg1 L DrsquoAcunto1 R Gregory-Clarke2

1TNO NL2BBC RampD UK

Abstract This paper reports on the latest developments in tiled streaming As an extension of HTTP adaptivestreaming it retains all the benefits of this streaming technology while adding the potential of interactionwhen UHDTV is consumed on mobile devices In particular we discuss the underlying principles and aspectssuch as multi-layer resolution scaling spatial segmentation and adaptive streaming Then we present insightsfrom a number of technology validation tests and demonstrations such as a live dance performance inManchester in May 2013 a training tool for professional skiers employing tiled streaming in Schladming hostof the Alpine Skiing World Championship 2013 and tests incorporating lsquoaugmented realityrsquo-style overlays inan athletics stadium in preparation for a trial at the 2014 Commonwealth Games Finally we report on thestatus of on-going standardization efforts in the MPEG-DASH ad-hoc group where tiled streaming isconsidered as a new feature referred to as Spatial Relationship Description

IntroductionThe momentum behind UHDTV is quickly gathering andUHDTV displays are entering the market at an increasingpace And although the production of UHD content islagging behind online content providers such as NetflixAmazon and YouTube have announced their plans forreleasing TV series and films in UHD formats such as 4KWith UHD format recommendations and requirementsemerging from ITU [1] and DVB (httpswwwdvborgnewsuhdtv-new-evidence-and-new-questions-for-dvb) withthe specification of a 4K Blu-ray format (httpwwwhollywoodreportercombehind-screences-as-ultra-hd-train-669587) and with several live 4K trials over existing broadcast(httpwwwbroadbandtvnewscom201405274k-smash-for-french-open) and CDN infrastructures (httpwwwiptv-newscom201405vienna-state-opera-streams-in-4k-with-elemental) the future for UHD (initially with 4Kpixels) looks bright For mobile devices the expectationsand benefits of UHD formats are less clear That is 4KUHD tablets and smartphones featuring limited screensizes are less-than-ideal candidates for displaying 4K UHDvideo And with an 8K UHDTV standard emerging the

discrepancy between native content resolution and screenrendering resolution will widen While it is possible toenable regular UHDTV experiences on tablets andsmartphones this experience can be enriched by(i) allowing end-users freely to extract a region-of-interestand navigate around the ultra-high resolution video and(ii) adding scalable lsquoaugmented realityrsquo-style overlays to thevideo Such an approach requires efficient delivery andmedia-aware network-based processing in order to supportmobile terminals and bandwidth limitations in the accessnetworks

An emerging technology referred to as rsquotiled streamingrsquoenables interaction with streaming video in such a way thatend-users can enjoy the full UHD resolution even if theirdevice is not capable of rendering and displaying the videoin its entirety As an extension of HTTP adaptivestreaming it retains all the benefits of this streamingtechnology while adding the potential of interaction whenconsuming UHDTV on mobile devices This paper reportson the latest developments in tiled streaming In particularwe (i) discuss the underlying principles and aspects suchas multi-layer resolution scaling spatial segmentation and


adaptive streaming and (ii) present insights from a numberof technology validation tests and demonstrations such as alive dance performance in Manchester a training tool forprofessional skiers employing tiled streaming inSchladming (host of the Alpine Skiing WorldChampionship in 2013) and tests incorporatinglsquoaugmented realityrsquo-style overlays in an athletics stadium inpreparation for a trial at the 2014 Commonwealth GamesWe further report on the status of on-going standardizationefforts in the MPEG-DASH ad-hoc group where tiledstreaming is considered as a new feature referred to asSpatial Relationship Description

Related workWith recent capturing systems for panoramic andomnidirectional UHD video new types of mediaexperiences are possible where end-users have the abilityfreely to choose their viewing direction and zooming levelMany different examples of such interactive video deliveryhave been demonstrated or deployed In the entertainmentsector companies like Immersive Media (httpimmersivemediacom) and Mativision (httpwwwmativisioncom) offer web streaming and mobile appsolutions to cover events with a 360-degree video cameraHowever such solutions rely on streaming or downloadinga complete spherical panorama to end-user devices wherethe final rendering of the interactive viewport takes placeIn an alternative approach KDDI (httpwwwtalkandroidcom22426-kddi-brings-zoom-enhance-for-hd-movie-streaming-on-smartphones) has shown a solution where allrendering takes place on the server side Here a low-powered and low-resolution mobile phone sends a spatialrequest to the server requiring the server to reframe andrescale the content accordingly before compression andstreaming to the end-user device Tiled streaming hasemerged as a scalable and bandwidth-efficient approach tointeractive UHD Initially proposed by Mavlankar [2] andfurther developed in [3ndash5] interactive UHD is enabled bya multi-layer tiling approach where the video is split intomultiple independently-encoded tiles which are re-stitchedin the end-user device

In addition to interactive video delivery panoramic andUHD video also provide a good opportunity for addinghotspots and overlays That is due to the static nature ofthe background additional and interactive overlay graphicscan be positioned in the world reference frame rather thanhaving to account for camera motion Such hotspots wereinitially incorporated into static panoramic systems such asQuicktime VR introduced in 1994 More recently systemsusing video panoramas have been developed for use insports broadcasting where real-time data relating to playertracking can be overlaid on the scene prior to selecting awindow for broadcast (httpwwwstatscompdfsSportVU_SonyDAVpdf) Approaches for adding interactiveoverlay graphics to conventional web video at the client sideare starting to appear (httpspopcornwebmakerorg) but

these are not yet generally being applied to interactivedelivery of panoramic and UHD video

Tiled streaming for zoomable videoZoomable video allows users to selectively zoom and pan intoregions of interest (ROI) within the video Such interactiontypically requires dynamic cropping of ROIs in the sourcevideo as well as unicast streaming of the cropped ROIsThe concept of tiled streaming addresses the limitations oftodayrsquos networks when streaming high resolution contentas well as allowing new applications in video streamingsuch as interactive panning and zooming Tiled videoprovides a better user experience than with predefined ordynamically cropped regions of interest Moreover itprovides better image quality for the selected region thanby simply enlarging pixel dimensions Finally the adaptiveversion of tiled streaming offers a new adaptationdimension after bandwidth resolution and quality ie theability at a given bandwidth to choose between full-framevideo in low quality and a spatial area in higher quality

Basic concepts of tiling

A tiled video can be obtained from a single video file orstream by partitioning each individual video frame intoindependently-encoded videos Tiles are thus defined as aspatial segmentation of the video content into a regulargrid of independent videos that can each be encoded anddecoded separately We denote the tiling scheme by MxNwhere M is the number of columns and N is the number ofrows of a regular grid of tiles See Figure 1 for twoexamples of regular tiling grids The tiling and subsequentseparate encoding of tiled videos leads to a reducedcompression performance due to reduced exploitation ofspatial correlation in the original video frame being limitedto tile boundaries This compression performance loss canbe reduced by using multiple resolution layers Eachadditional layer originates from a lower resolution versionof the original video frame tiled into a grid with fewertiles If the tiling is small enough (such as thumbnails) thebit-rate overhead of using another resolution layer isaffordable This multi-resolution tiling increases the qualityof user-defined zooming factors on tiles Once a user

Figure 1 Example of 2x2 (left) and 4x4 (right) tiling gridsDepending on the tile size a single ROI overlaps multipletiles and thus requires multiple tiles for reconstruction



zooms into a region of the content the system will providethe highest resolution tiles that are included in therequested region

Optimizing performance on low-powereddevices with overlapping tiling

Figure 2 provides an example of an overlapping tiled grid asused in [6] Such a tiling is beneficial for mobile devices thatare equipped with a single hardware decoder only In thiscase the total number of tiles required to reconstruct therequested ROI can be reduced to one while the ROI sizeand total number of tiles remain approximately the sameThe effectiveness of overlapping tiles in reducing thenumber of tiles required to reconstruct a given ROI isdetermined by the overlapping factor which gives therelative overlap (per planar direction) of a particular tile inrelation to its size Choosing a larger overlapping factorresults in larger overlapping areas and thus in fewer tilesbeing required to reconstruct a given ROI The downsideof this overlap is that these redundant pixels result in alarger amount of data to be stored on the server side Alsooverlapping tiles result in heterogeneous tile sizes

Tiled adaptive streamingAs shown in Figure 3 spatial segmentation can becomplemented with the temporal segmentation of HTTPAdaptive Streaming (HAS) The scalability properties ofHAS enable zoomable video to be available to a largenumber of users thanks to efficient bandwidth utilisationcacheability and simpler inter-tile synchronization Tiledstreaming can be integrated within HAS by having eachvideo tile individually encoded and then temporallysegmented according to any of the common HAS solutions(eg MPEG DASH [7] or Apple HLS (HTTP LiveStreaming see httptoolsietforghtmldraft-pantos-http-live-streaming-13) This leads to a form of tiled adaptivestreaming where all tiles are temporally aligned such thatsegments from different tiles can be recombined to createthe reassembled picture An advantage of using HAS forthe delivery of spatial tiles is that the inherent time-segmentation makes it relatively easy to resynchronisedifferent spatial tiles when recombining tiles into a singlepicture or frame As long as the time segmentation processmakes sure that time segments between different spatialtiles have exactly the same length the relative position of aframe within a time segment can be used as a measure for

Figure 2 (left) Tiling scheme with overlapping tiles (2x2 grid) and an overlap factor of one third (right) Overlapped tilingscheme with an overlap factor of one fourth

Figure 3 With tiled HAS a video is tiled in a certain grid Each tile is encoded and segmented using HAS segments In thisexample the grid is 2 by 2



the position of that frame within the overall timeline Forexample frame number n within time segment s of tile A issynchronised with frame number n within time segment sof tile B On the client side timestamps provided by thesegment container can be used to ensure perfectsynchronisation between the segments that make up thefinal viewport to be rendered on the screen of the end user

Applications of tiled streaming technology

In this section we present insights from a number ofrecent and planned technology validation tests anddemonstrations such as (i) the FascinatE system usedduring a live dance performance in Manchester (ii) theiCaCoT system used as a training tool for professionalskiers in Schladming February to April 2014 and (iii)preparations for tests including lsquoaugmented realityrsquo-styleoverlays in an athletics stadium for the 2014Commonwealth Games

Fascinate system

Tiled streaming was employed for the distribution ofpanoramic video sequences in the context of the European-funded project FascinatE (Format-Agnostic SCript-basedINterAcTive Experience see httpwwwfascinate-projecteu) A live system was demonstrated during a dance

performance in Manchester in May 2013 see Figure 4 and 5Here a multi-camera audio-visual scene representationincluding both panoramic 6K and regular 1080p videocontent was spatially tiled and temporally segmented Fortiling a dyadic tiling approach was used ranging from 1x1to 8x8 tiling grids resulting in a multi-resolution set ofpanoramic video layers with the original resolution of 6976by 1920 pixels as a base layer

iCaCoT System

Tiled streaming was further incorporated into the iCaCoT(Interactive CAmera-based COaching and Trainingsee httpwwwexperimediaeu20140220icacot) trainingapplication in the context of the European-funded projectEXPERIMEDIA (Experiments in live social andnetworked media experiences see httpwwwexperimediaeu)Here the goal was to provide ski coaches in Schladmingwith a tablet application through which they could providetheir trainees with real-time feedback This was achievedusing a combination of a set of static high-resolutioncameras along the ski-slope tiled streaming and advancedtrick-play and drawing features see Figure 6 Incollaboration with Schladming2030 the Austrian venuepartner we performed several experiments with a live4K-tiled streaming system A significant challenge was tocope with the particular conditions of the experiment

Figure 4 Panoramic 6K image of the FascinatE live demonstration based around the performance of rsquoDeeper than all rosesrsquoa composition from Stephen Davismoon featuring rock band BearsBears and live performance artists Joseph Lau andShona Roberts

Figure 5 The live performance was shown in a separate room delivered via tiled streaming The presenter could navigatethrough the video panorama using a tablet Additional footage from earlier recordings was also shown on tablets



location as the material had to be protected from extremeweather conditions in terms of temperature humidity andso on As the on-site connectivity to the open Internet ruledout any off-site video processing such as tiling and encodingthe overall system had to be stand-alone and compact

For tiling of 4K video at 24 fps in real-time a software-based approach was not a viable solution Therefore ahardware-accelerated pipeline based on the Intel MediaSDK (Intel Media Software Development Kit see httpsoftwareintelcomen-usvcsourcetoolsmedia-sdk-clients)was used This pipeline running on a single machine is ableto decode the original panorama to produce the tiles and toindependently encode them in H264 at 24 fps In practicethe pipeline can handle in parallel more than 10 tiles atroughly 1920x1080 resolution

Commonwealth games 2014 system

These experiments aim to verify that navigation around ahigh resolution video using tiled streaming will contributeto a higher sense of interaction and engagement amongstusers particularly in the case of large-scale event-basedprogramming This form of content often contains severaldistinct regions of interest and a number of differentthings that a user may want to look at in more detailHence we hypothesise that the user experience could befurther enhanced through the use of overlaid interactivegraphics which provide extra information about the scene

An example of such an application is an athletics eventwhich typically features several different track and field events

occurring simultaneously As well as being able to pan andzoom around the scene as they wish the user can also bepresented with data which offers more detail about what theyare looking at This could include the locations and times ofthe sports taking place on the day in question the names ofthe athletes that are visible the current height of the highjump bar and so forth The intention is to provide the typeof rich data that a user would typically be interested inanyway but that they would ordinarily have found eitherfrom burnt-in graphics provided by a broadcaster or else aself-initiated search Presenting this information as optionaloverlays in this way allows the user to access the level ofdetail they want when they want it without having to leavethe application We plan to test interactive overlays onpanoramic video as part of a closed trial of a prototypesystem known as the Venue Explorer at the CommonwealthGames in Glasgow in July-August 2014 The plan is tocapture a wide-angle view of the athletics stadium using a 4kcamera encode this as a set of overlapping tiles and streamthese to an HTML5-based client using MPEG-DASHData relating to sports events (including live updates) will beused to render interactive overlays in the client according tooptions selected by the viewer Initial work on the systemused video captured at the London Anniversary Games insummer 2013 and is shown in Figure 7

Standardisation of tiled streamingMPEG-DASH (ISOIEC 23009) is the adaptive streamingtechnology standardised by MPEG After having published afirst edition of the standard [7] MPEG experts are now

Figure 6 Interactive camera-based coaching and training using tiled streaming for video navigation Several applicationscreenshots with the third screenshot showing user-drawn markers helping to provide the skier useful feedback

Figure 7 A possible user interface for the Venue Explorer system



aiming to extend the original scope of adaptive streaming tonew lsquouse-casesrsquo Tiled streaming use-cases and their relevancywere presented to the MPEG-DASH working group duringthe 104th MPEG meeting in April 2013 There MPEG-DASH experts acknowledged the usefulness of such use-cases and agreed to start a so-called Core Experiment Thegoal of this Core Experiment was to steer the group efforttowards a technical solution which would permit these tiledstreaming use-cases The discussions within the CoreExperiment reached a consensus among the group at the107th MPEG meeting last January Consequently MPEGhas initiated the publication process of this new featurewhich should be completed during 2015

Conceptually this new feature called Spatial RelationshipDescription (SRD) allows an author of the MPEG-DASH media presentation description (MPD) to describehow the various tiles are spatially related with each otherThis description handles both intra and inter layerrelationships Thus far the standard only allows for anAdaptationSet to define perceptually-equivalent contentTherefore describing different tiles under the sameAdaptationSet would violate this rule With the SRDfeature the concept of a tile as defined in this paper ismapped onto the AdaptationSet element of the MPD It isalso important to note that this new mapping decouplesthe tile concept from a particular video To offer backwardscompatibility it remains possible to benefit from theregular properties of AdaptationSets namely the availabilityof several Representations in different bit-rates codecsresolutions and so on In practice the new MPEG-DASHSRD feature will specify a set of parameters in order todescribe tiles with respect to a common reference spaceThese parameters are (xy) respectively horizontal andvertical positions of the tile (wh) respectively widthand height of the tile and (WH) respectively width andheight of the reference space All these values are expressedin an arbitrary unit as chosen by the MPD author

Future workIn subsequent developments we aim to improve the livetiling process such that it can handle a range of UHDformats including higher resolutions and frame ratesIncorporation of multi-sensor systems will also beconsidered In the near future we aim to produce live 4Kfootage and to deliver this via tiled streaming to end-usersin a large-scale user trial This allows us to determine theimpact of live 4K-tiled streaming on a regular productionenvironment and to measure the effects of tiled streamingon bandwidth and latency in a regular content deliverysetting Furthermore the ways in which users interact with

panoramic UHD video and overlays will be studied in aseries of user trials which will inform future developmentsIt is also planned to evaluate the use of interactivepanoramic video with overlays in other application scenarios

AcknowledgmentsThe research leading to these results has received fundingfrom the European Unionrsquos Seventh FrameworkProgramme (FP72007-2013) under grant agreements no248138 and no 287966

References

[1] ITU-R Recommendation BT2020 ldquoParameter valuesfor ultra-high definition television systems for productionand international programme exchangerdquo August 2012

[2] MAVLANKAR A AGRAWAL P PANG D HALAWA S CHEUNG N-MGIROD B lsquoAn Interactive Region-Of-Interest VideoStreaming System For Online Lecture Viewingrsquo SpecialSession on Advanced Interactive Multimedia StreamingProc of 18th International Packet Video Workshop (PV)Hong Kong December 2010

[3] KHIEM N RAVINDRA G CARLIER A OOI WT lsquoSupportingZoomable Video Streams via Dynamic Region-of-InterestCroppingrdquo inrsquo Proc of 1st ACM Multimedia SystemsConf Scottsdale Arizona 22 ndash 23 February 2010pp 259ndash270

[4] BRANDENBURG R VAN NIAMUT O PRINS M STOKKING HlsquoSpatial Segmentation For Immersive Media Deliveryrsquo InProc of 15th Int Conf on Intelligence in Next GenerationNetworks (ICIN) Berlin Germany 4 ndash 7 October 2011pp 151ndash156

[5] QUAX P ISSARIS P VANMONTFORT W LAMOTTE W lsquoEvaluationof Distribution of Panoramic Video Sequences in theeXplorative Television Projectrsquo NOSSDAVrsquo12 TorontoCanada 7ndash8 June 2012

[6] PANG D HALAWA S CHEUNG N-M GIROD B lsquoClassX MobileRegion-of-Interest Video Streaming to Mobile Deviceswith Multi-Touch Interactionrsquo In Proc of MMrsquo11Scottsdale Arizona USA November 28ndashDecember 1 2011

[7] MPEG ldquoISOIEC DIS 23009-1 Information Technology ndashDynamic adaptive streaming over HTTP (DASH) ndash Part 1Media presentation description and segment formatsrdquoJanuary 5 2012



ISSN 2041-5923

High frame-rate television sampling theorythe human visual system and why theNyquist-Shannon theorem doesnrsquot applyKC NolandBBC Research and Development UK

Abstract In this paper an analysis of video motion portrayal in terms of classical sampling theory is presentedthat uses measurements of the human visual system to determine the highest temporal frequency that theframe-rate should be able to support A discussion of the limitations of a traditional sampling theoryapproach is also presented Additional analysis is given of the effects of camera shuttering and displayprocessing which are essential parts of any real system This is the first time that such analysis has beencarried out in the context of ultra-high definition television

IntroductionUltra-high definition (UHD) television standards are in theprocess of being formalised and there is currently muchdiscussion about suitable parameter choices Higher frame-rates have received particular attention with subjective testscarried out by the European Broadcasting Union [1] havingdemonstrated that simply increasing spatial resolutionproduces only a small quality improvement and that using ahigher frame-rate can significantly improve the perceived quality

Digital video capture is a three-dimensional samplingprocess to which it is possible (although not necessarilydesirable) to apply traditional sampling theory principlesThe frame-rate relates to the time dimension After areview of video motion artefacts and related work thisarticle presents an explanation of how sampling theory canbe used to find a suitable frame-rate assuming that nostrobing can be tolerated This requires a limit for thehighest temporal frequency in the signal A model of thehuman spatio-temporal contrast sensitivity function (CSF)is used to find the highest temporal frequency needed

Then follows a discussion on the effects of eye movementswhich mean that standard sampling theory is insufficient tomodel motion artefacts Measurements of eye trackingspeeds are used to model increased blur perception duringtracking and then applied to the frame-rate calculations

Another consequence of eye tracking is that the aliasingartefacts that can cause strobing are masked Theperception of what would traditionally be called aliasing isstill poorly understood and requires extensive furtherresearch One way to balance blur and strobing is to controlthe camera shuttering and display hold times and thisarticle also presents a theoretical analysis of these effects

Motion artefacts in videoThere are three important artefacts that are affected by theframe-rate large area flicker which is the result of thewhole screen being refreshed at the frame-rate visible ondisplays that have some blanking between frames motionblur caused by objects moving while the camera shutter isopen and strobing caused by having too few frames persecond for the motion to appear smooth Motion blur andstrobing can be traded against each other by varying thetime that the camera shutter remains open for each framecapture but the only way of improving them both togetheris to increase the frame-rate

The perception of motion is also affected by eyemovements When the eye follows a moving object duringeye tracking the relative speed between the eye and theobject is reduced and hence the image moves more slowlyacross the retina than when the eye is still This increasessensitivity to motion blur but decreases sensitivity to



strobing For displays with a long on-time additional motionblur is created in the eye as it moves while the displayed imageremains still for a frame period This is sometimes calledldquoretinal sliprdquo

Previous researchThe ITU-R Report BT2246 [2] The present state of ultrahigh definition television contains an overview of subjectiveexperiments carried out by Japanese Broadcaster NHK onframe-rates for television The results show that around 80frames per second (fps) were needed to eliminate flickerwith 100 degrees field of view much lower than the frame-rates required to reduce motion blur and strobing toacceptable levels For motion blur it was found that thecamera shutter length should be less than about 3 msFurther experiments showed great improvements in theperceived strobing at 120 fps over 60 fps with additionalperceptible improvements at 240 fps Strobing was alsoshown to decrease with a longer camera shutter Overallsubjective ratings confirm that 120 fps offers a bigimprovement in subjective quality over 60 fps and 240 fpsis better still

Kuroki et al [3] show that in 24 frames per secondprogressive and 30 frames per second interlaced formatsvisual acuity is much reduced in comparison to directviewing as soon as any movement occurs This is clearevidence that the frame-rates tested are insufficient for atransparent system Subjects also rated blur and jerkinessfor material at frame-rates from 62 to 500 fps and adramatic reduction in both artefacts was reported at 125 fpsover 62 fps then flattening off above 250 fps Jerkiness wasalso rated worse with short shutter times Theseexperiments independently verify the results presented inITU-R BT2246 [2]

In much earlier experiments with high definition (HD)television Stone [4] demonstrated significant perceivedimprovements when increasing the frame-rate from 50 to80 fps He reports that although reducing the camerashutter time to 50 also improved the quality of motionthis approach caused severe problems when converting toother frame-rates due to the strong temporal aliases in thesignal More recently Armstrong et al [5] give descriptionsof the arguments for higher frame-rates and reportobservations of significant quality improvements in materialdisplayed at 100 fps over conventional rates

Tonge [6] and Watson [7] both took a more analyticalapproach Tonge suggested that a frame-rate of up to 2000fps would be needed for television pictures to appearexactly equivalent to continuous motion based onmeasured limits of dynamic resolution He also noted thatwith eye tracking the rate can be reduced to the criticalflicker frequency Watson analysed visibility of artefactsaccording to whether they appear in the ldquowindow of

visibilityrdquo which he defined as a threshold on the CSFmeasured by Robson [8]

Applying sampling theory principlesto video frame-rateClassical sampling theory teaches that the samplingfrequency must be at least twice the highest frequency inthe signal This is a fundamental principle of digital signalprocessing and is usually attributed to Nyquist [9] andShannon [10] When applied to video capture thistranslates to a frame-rate of at least double the highesttemporal frequencymdashthe rate at which an individual pixelchanges its brightness most often as a result of an objectmoving past that pixel position The rate at which thebrightness changes depends on the amount of spatial detailin the object and its speed This relationship can beformalised as

t = rn (1)

where t is the temporal frequency r is the spatial frequencyand n the velocity [6] A flashing object can also create atemporal frequency in which case the flash frequencyshould also be incorporated into equation 1 but this occursrelatively rarely in real scenes

The frame-rate must therefore be at least 2t and can becalculated for any r and n The pixel format enforces anupper limit on the spatial detail that can occur but it ismuch more difficult to determine the highest velocity thatwe want to represent To find a suitable upper limit onvelocity the next section proposes using measurements ofthe human visual system to perceptually match the spatialand temporal resolution of a format First it is assumed thatthe viewer has a fixed gaze then a model of eye tracking isincorporated

Human contrast sensitivity functionThe human CSF models sensitivity to different spatialfrequencies moving past at different velocities Our abilityto resolve spatial detail drops both with higher spatialfrequency and with higher velocity Laird et al [11] fit amodel of the CSF originally proposed by Kelly [12] totheir own experimental data as shown in figure 1 Subjectsmaintained a fixed gaze during the experiments so the datadoes not take eye tracking into account

Using the contrast sensitivityfunction to find a limit on velocitySince it is not possible to put an upper limit on the velocitiesthat may occur in a scene it is proposed instead to restrict therepresentable velocities to those that would cause loss ofspatial detail due to motion blur in the signal to be noworse than loss of spatial detail caused by limiting thespatial resolution Our ability to resolve the highest spatial



frequency of a particular format is represented by the heightof the CSF at that spatial frequency and at a velocity ofzero indicated by the thick vertical line in figure 1 whichis at 1920 cycles per picture width In order to be able torepresent all moving objects that we can resolve just as wellas a static object of the highest spatial frequency it isnecessary to follow one contour line in the CSF asillustrated by the thick curved line in figure 1 Each pointalong the line has an associated spatial frequency andvelocity from which a frame-rate can be calculated usingequation 1 The highest of these frame-rates is proposed asthe overall required rate

The CSF data is specified in units of cycles per degreesubtended at the retina and degrees per second at the retinaTo use the more convenient units of cycles per picture widthand picture widths per second it is necessary to take accountof the viewing distance Full details of the conversion areprovided by Noland [13] Figure 1 is plotted using the newunits for a distance of 15 times the screen height (15 H)

Figure 2 shows the calculated frame-rates for three viewingdistances With 720 horizontal pixels viewed at 3H therequired frame-rate is a little below 50 fps This matchesthe standard definition format in the UK which has aninterlaced field-rate of 50 fields per second The calculatedframe-rate then increases approximately linearly with thenumber of horizontal pixels up to around 140 fps for 1920horizontal pixels Beyond this point there are no morecontrast sensitivity measurements so it is not valid toextrapolate the lines shown At viewing distances of 15 Hand 075 H that are recommended for UHD with 3840and 7680 horizontal pixels respectively the results suggestthat without eye tracking 140 fps would also be suitableThis is because the recommended viewing distances are alldesigned to put the pixel structure at the limit of humanspatial acuity so the same frame-rate is needed to matchthe spatial resolution Viewing from further away brings nobenefit in spatial resolution beyond the limit of acuity with

ideal Nyquist-Shannon reconstruction at the displayFurther discussion of this is given in [13]

Eye tracking modelThe CSF model is based on data collected with a fixed gazebut under normal viewing conditions the eye is free to followmoving objects which has the effect of improving our abilityto resolve spatial detail This implies that objects can movemuch faster before we are unable to resolve their spatial detailthan when the eye is fixed and hence a higher frame-ratewould be required to keep motion blur to an acceptable level

A model of human eye tracking based on experimentaldata was incorporated into the CSF model Daly [14pp 185ndash187] describes how on average humans trackmotion at about 82 of the object speed and finds amaximum tracking speed of about 80 degrees per secondThe model is shown in figure 3

To adapt the frame-rate calculations for eye trackingall velocities in the CSF were scaled up by1(1 2 082) asymp 56 to give the speed of the tracked object

Figure 1 Contrast sensitivity function model for a viewingdistance of 15 H and a fixed gaze at the centre of thescreen The thick line shows the threshold of spatio-temporal frequencies to be represented to perceptuallymatch the spatial resolution of UHD-1

Figure 2 Frame-rate required to perceptually match variouscommon horizontal resolution values with viewingdistances of 075 H 15 H and 3 H and a fixed gazeassuming no strobing can be tolerated

Figure 3 Eye tracking model



that would create an image moving at the original unscaledspeed on the retina during tracking An upper limit of 80degrees per second was also imposed on the tracking

velocity The corresponding maximum trackable objectvelocity is (80082) frac14 97 degrees per second

Figure 4 shows the calculated frame-rates according to thecontrast sensitivity model with incorporated eye trackingThe figures demonstrate just how important an effect eyetracking is in the discussion of motion portrayal and frame-rates Frame-rates of up to 700 fps are now coming intoconsideration

Why the Nyquist-Shannon theoremdoesnrsquot applyBefore dwelling on the clearly impracticable figure of 700 fpsit is important to consider the assumptions made to reach itThe Nyquist-Shannon theorem assumes that no aliasing canbe tolerated However this is not necessarily the case In facttelevision systems have always been temporally undersampled[15] The reason this does not look more objectionable is thatthe moving eye is an important part of video signalreconstruction and is not taken into account by traditionalsampling theory which requires no aliasing only from the

Figure 4 Frame-rate required to perceptually match variouscommon horizontal resolution values with a viewingdistance of 075H 15H and 3H during eye trackingassuming no strobing can be tolerated

Figure 5 Illustration of camera shutter effectsa Nyquist-Shannon samplingb Nyquist-Shannon sampling with alias and ideal reconstruction filterc 100 camera shutterd 100 camera shutter with alias and ideal reconstruction filtere 50 camera shutterf 50 camera shutter with alias and ideal reconstruction filter



perspective of a fixed position on the screen Eye tracking canbe interpreted as a kind of motion-compensatedinterpolation where parts of the signal that wouldtraditionally be called aliases from the perspective of a fixedposition on the screen are correctly unwrapped by the eyeto be perceived at the right speed This means that fortracked motion to look smooth it may only be necessaryfor the frame-rate to be above the critical flicker frequency

However the eye can only track one velocity at a time andany untracked motion will still be subject to strobing effectsLittle is currently known about how the visibility of strobingartefacts varies with different kinds of motion frame-ratescamera shutter times and display apertures but thatknowledge is essential to make a complete judgement aboutsuitable frame-rates If signal aliases are to be eliminatedthe calculations suggest that a frame-rate of 700 fps isneeded to maintain resolution in tracked motion Howeverif it is found that some strobing can be tolerated and it isassumed that aliases are the sole cause of strobing artefactsthe strict criterion of having no aliasing in the signal couldbe relaxed and hence the required frame-rate could bereduced Better understanding of strobing perception wouldalso allow an informed judgement to be made regardingthe best balance between blur and strobing for a morerealistic frame-rate such as 100 fps

The effects of camera Shutteringand display sample-and-holdOne way to balance blur and strobing effects is to vary theamount of time for which the camera shutter is open Thelonger the shutter time the more blurred the image

Conversely the shorter the shutter time the further an objectwill have moved between one frame and the next and so themore strobing will be present Camera shuttering can beinterpreted as a filtering operation the light entering thecamera is integrated for the time that the shutter is openwhich is equivalent to convolution with a rectangle functionprior to sampling Viewed in the frequency domain thiscorresponds to a multiplication of the signal spectrum with asinc function as illustrated in figure 5

Figure 5a shows the unrealistic case that was assumed bythe Nyquist-Shannon analysis where a so-called ldquobrickwallrdquo anti-alias filter passes all content below half thesampling frequency and removes everything at higherfrequencies Figure 5b shows the alias in the sampledsignal which does not overlap at all with the basebandWhen instead a 100 camera shutter is used as the anti-alias filter as in figure 5c the null of the frequency-domainsinc function appears at the sampling frequency Upperpassband frequencies are slightly attenuated and stopbandfrequencies whilst attenuated are not completely removedHowever due to the position of the null the alias (figure5d) has a very low level around zero frequency and hencestrobing is not severe With a 50 shutter (figure 5e) thefrequency-domain sinc function has its null at twice thesampling frequency This causes much less signalattenuation than a 100 shutter and hence much less blurbut also means that the notch in the alias (figure 5f) is nolonger conveniently positioned at DC and so strobingbecomes highly visible in untracked motion

The reconstruction filter has been shown as an idealisedldquobrick wallrdquo filter in figure 5 but reconstruction filtering is

Figure 6 Illustration of display effectsa 100 camera shutter followed by sample-and-hold displayb 100 camera shutter followed by impulsive displayc 50 camera shutter followed by sample-and-hold displayd 50 camera shutter followed by impulsive display



in reality provided by the display in conjunction with thehuman visual system It is therefore instructive toadditionally consider the effects of different kinds oftemporal display response on the signal

A conventional liquid crystal display (LCD) will haveclose to a 100 on-time Viewed as a filtering operationthis multiplies the sampled spectrum by a sinc functionas illustrated in figure 6 for a 100 camera shutter(figure 6a) and a 50 camera shutter (figure 6c) In bothcases the display sample-and-hold causes additionalattenuation resulting in more motion blur Aliases in thesignal due to the capture process are also attenuatedalthough not completely removed Strobing visibility isfurther masked in sample-and-hold displays when the eyeis moving by the addition of motion blur as a result ofldquoretinal sliprdquo

If instead an impulsive display response is used where thedisplay is on for a very short proportion of the frame periodthe display essentially leaves the signal spectrum unchangedMotion blur will not be made worse by the display in thiscase but it cannot be removed For a long camera shutter(figure 6b) blur will still be present in the signal and fora short camera shutter (figure 6d) the strobing will behighly visible in non-tracked motion If blur and strobingare well-balanced during capture it would be preferable touse a display with a short hold time since this would alterthe signal least However at 50 or 60 fps flicker is stillvisible on large impulsive displays so additional processingsuch as motion-compensated frame insertion is neededwhich adds its own motion artefacts For a frame-rateabove the critical flicker frequency impulsive displays are anattractive option

ConclusionAn analysis of video sampling in terms of classical Nyquist-Shannon theory has been presented and used togetherwith models of the human contrast sensitivity function andof eye tracking to suggest a frame-rate that perceptuallymatches a given spatial resolution in the sense that spatialdetail is not lost disproportionately when objects moveThe frame-rates proposed should be regarded asapproximate due to their reliance on experimental averagesMore significantly the analysis does not take into accountany tolerance we may have to strobing effects It is likelythat lower frame-rates will be acceptable if a degree ofaliasing in the signal can be permitted but to confirm thisfurther research into the visibility of strobing artefacts isneeded

One method of controlling the balance between blurand strobing is to vary the camera shutter and displayaperture times An analysis of these effects illustratedthat there will always be some attenuation of wantedsignal with a finite aperture time so some temporaloversampling may be desirable A shorter camera shutter

time can reduce the attenuation but strongly boosts thelevel of any aliases which cause strobing especially innon-tracked motion A better understanding of theseverity of strobing artefacts is also essential to adecision on suitable camera and display processing for agiven frame-rate

References

[1] EBU 2013 Subjective testing confirms importance offrame-rate for UHDTV httpstechebuchnewssubjective-testing-confirms-importance-o-07aug13

[2] ITU-R 2012 The present state of ultra high definitiontelevision ITU-R BT2246ndash1

[3] KUROKI Y NISHI T KOBAYASHI S OYAIZU H YOSHIMURA S lsquoApsychophysical study of improvements in motion-imagequality by using high frame-ratesrsquo J Soc InformationDisplay 2007 15 (1) pp 61ndash68

[4] STONE MA lsquoA variable-standards camera and its useto investigate the parameters of an HDTV studiostandardrsquo BBC Research Dept Report RD 198614Report 1986

[5] ARMSTRONG M FLYNN D HAMMOND M JOLLY S SALMON RlsquoHigh frame-rate televisionrsquo BBC Research White PaperWHP 2008 p 169

[6] TONGE GJ lsquoTime-sampled motion portrayalrsquoProc 2nd Intl Conf Image Processing and itsApplications 1986

[7] WATSON AB lsquoHigh frame-rates and human visionA view through the window of visibilityrsquo SMPTE MotionImaging J 2013 122 (2) pp 18ndash32

[8] ROBSON JG lsquoSpatial and temporal contrast-sensitivityfunctions of the visual systemrsquo J Opt Soc Am 1966 56(8) pp 1141ndash1142

[9] NYQUIST H lsquoCertain topics in telegraph transmissiontheoryrsquo Trans American Institute of Electrical Engineers1928 47 (2) pp 617ndash644

[10] SHANNON CE lsquoA Mathematical Theory ofCommunicationrsquo Bell System Technical J 1948 27 (34)pp 379 to 423 623ndash656

[11] LAIRD J ROSEN M PELZ J MONTAG E DALY S lsquoSpatio-velocityCSF as a function of retinal velocity usingunstabilized stimulirsquo Proc SPIE-ISampT Electronic Imaging2006 6057

[12] KELLY DH lsquoMotion and vision II Stabilized spatio-temporal threshold surfacersquo J Opt Soc Am 1979 69 (10)



[13] NOLAND KC lsquoThe Application of Sampling Theory toTelevision Frame-rate Requirementsrsquo BBC RampD WhitePaper WHP 2014 Vol 282

[14] DALY S lsquoEngineering observations from spatiovelocityand spatiotemporal visual modelsrsquo in Christian J van den

Branden Lambrecht(editor) Vision Models andApplications to Image and Video Processing 2001 KluwerAcademic Publishers

[15] DREWERY JO lsquoTV-the grand illusionrsquo IEE Colloq MotionReproduction in Television 1995 pp 11ndash17




ISSN 2041-5923

Development of 8K-UHDTV system with wide-colour gamut and 120-Hz frame frequencyT Soeno T Yasue K Kitamura K Masaoka Y KusakabeH Shimamoto T Yamashita Y Nishida M SugawaraNHK (Japan Broadcasting Corporation) Japan

Abstract NHK has been researching Super Hi-Vision (SHV) as a next-generation broadcasting system SHV aims atan ultra-high definition television (UHDTV) system with a 7680 times 4320 resolution (8K resolution) 120-Hz framefrequency and wide-colour gamut which we refer as lsquofull-specification SHVrsquo although its RampD was initiated withlimited parameter values owing to the available technologies

In this paper we discuss the equipment developed for a full-specification SHV system Utilising CMOS imagesensors for capturing 8K images at 120 Hz and a prism designed for the wide gamut we produced the worldrsquosfirst full-specification SHV camera system A new signal interface for studio equipment was developed totransmit the full-specification SHV signal with a data-rate of approximately 144 Gbits using a single opticalcable In addition we developed an 8K120Hz liquid crystal display and 8K120Hz wide-colour gamut laserprojector We used these pieces of equipment to produce full-specification SHV video content and confirmedthe improvements in the image quality of 8K resolution 120-Hz frame frequency and wide-gamut colorimetry

IntroductionWe have been researching and developing an extremelyhigh-resolution video system called Super Hi-Vision (SHV)which is the ultimate two-dimensional television The videoparameters of UHDTV systems have been standardised inRecommendation ITU-R BT2020 [1] SMPTE ST 2036-1 [2] and ARIB STD-B56 [3] Among these parametersSHV aims at a system with a 7680 times 4320 resolution (8Kresolution) 120-Hz frame frequency 12-bit coding formatand wide-colour gamut We refer to such a system as lsquofull-specification SHVrsquo [4] although its RampD was initiated withlimited parameter values owing to the available technologiesThe realisation of full-specification SHV requires theevaluation of moving images and the development ofequipment such as a camera an interface and a displayThis paper reports on the development of such a system

First we describe the worldrsquos first full-specification SHVcamera system For constructing this camera we developeda CMOS image sensor (which is the key device of acamera system) for capturing 8K images at 120 Hz and a

prism designed for the wide-colour gamut Next wedescribe a new interface which is based on multilink10-Gbits streams which can transmit a full-specificationSHV signal with a data-rate of approximately 144 Gbitsusing a single optical cable Then we explain monitoringequipment for the UHDTV signals We developed an 8Kwide-colour gamut laser projector a colour-gamut monitorand a liquid crystal display We produced SHV content atfull-specification using the aforementioned equipment andconfirmed the improvements in the image quality of 8Kresolution 120-Hz frame frequency and wide-colourgamut colorimetry

Camera systemCamera design

With the aim of producing a full-specification SHV camerawe set the following requirements

dagger The camera shall be of a size which promotes mobility andoperability for indoor and outdoor shooting



dagger The camera shall capture wide-colour gamut images

dagger The camera shall capture 8K resolution (7680 times 4320pixels) images with a frame frequency of 120 Hz

dagger The camera system shall maintain the same sensitivity andsignal-to-noise ratio (SNR) as conventional SHV camerasystem [5]

dagger The system shall perform signal processing in real time

First we developed a CMOS image sensor which couldcapture 8K resolution images at a frame frequency of120 Hz to achieve the above requirements Table 1 lists thespecifications of the developed image sensor

Next we designed the colorimetry in the camera Thecolour gamut of the SHV as specified in Rec ITU-R

BT2020 is wider than that of HDTV as specified in RecITU-R BT709 [6] The values of the primaries and whitepoints specified in Rec BT2020 and Rec BT709 arelisted in Table 2 Figure 1(a) shows the ideal cameraspectral sensitivity curves for rec BT2020 primaries Thecurves are not physically realisable because of their negativesensitivities Therefore we designed practical spectralsensitivity curves composed of only the positive main lobesof the ideal curves by eliminating the second positive lobeand negative regions [7] as shown in Figure 1(b) Thecamera system applies a 3 times 3 linear matrix (LM) toapproximate the effect of the negative regions

The block diagram of the camera system is shown inFigure 2 The camera head consists of three CMOS imagesensors on headboards for each colour (red green andblue) the wide-gamut prism described above and anoptical transmission unit Dense wavelength divisionmultiplexing (DWDM) is used for signal transmissionbetween the camera head and the camera control unit(CCU) because we would prefer to use the same cables forconventional HDTV cameras to transmit the SHV signalThe image data are transformed into sixteen 10-Gbitsstreams and then output to the CCU The CCU performssignal processing including fixed-pattern-noise (FPN)cancellation colour correction by using a linear matrixsignal level control gamma correction and detailenhancement in real time After that the output signal is

Table 1 Image sensor specifications

Image size 25 mm (H) times 14 mm (V)

Optical image format 17 in

Active pixels 7680 (H) times 4320 (V)

Pixel size 32 mm (H) times 32 mm (V)

Frame frequency 120 Hz

Table 2 Comparison colour primaries and white point between Rec ITU-R BT 2020 and Rec ITU-R BT709

Rec BT2020 RecBT709

CIE x CIE y CIE x CIE y

Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060

Reference white (D65) 03127 03290 03127 03290

Figure 1 Spectral sensitivity for the SHV wide-colour gamut (a) ideal curves and (b) curves composed of the positive mainlobes of the curves with ideal spectral sensitivity



formatted into a new interface described in the followingsection Figure 3 shows the appearance of the camera head

Spatial resolution and SNR

An example of a reproduced image is shown in Figure 4 Weconstructed a chart corresponding to one-fourth of the entirescreen In the image a modulation response at 2000 TVlines which corresponds to 4000 TV lines by conversionwas observed We also measured the modulation transferfunction (MTF) of the luminance component (Y) by usingthe lsquoslanted edgersquo method with the ISO-12233 resolutionchart [8] Figure 5 shows the characteristics of themeasured MTF in the vertical direction The calculatedpoints denote the theoretical MTF calculated bymultiplying the respective component MTFs of the system

Figure 2 Block diagram of the camera system

Figure 3 Appearance of the camera head

Figure 4 (a) Example of reproduced image (7680 times 4320 px) (b) Enlarged view of the image corresponding to the red circlein (a) to measure around 2000 TV lines



the ideal aberration-free-lens MTF and the image sensorMTF [9] when the f-number of the lens is set to 40 Wecan observe that the MTF at the Nyquist frequency isapproximately 20 From the resolution measurementpresented above this camera has sufficient ability to capture8K resolution images

Measurement of the camerarsquos SNR revealed a value of50 dB (sensitivity setting luminance level is 2000 lx lensaperture sets at F 56) which is better than that of aconventional SHV camera system [5] despite having aframe frequency which is twice as high

Capturing moving objects

We captured both a moving object and a one-frame imagewhile the camera was operated at frame frequencies of120 Hz and 60 Hz (Figure 6) The motion blur of theimage with a frame frequency of 120 Hz was less than thatwith a 60-Hz frame frequency From the results weconfirmed that the frame frequency of 120 Hz improves theimage quality and the camera system can capture images at120 Hz

Colour reproducibility

The colour-reproduction characteristics of the camerasystem were measured using a 24-patch transmission

colour checker chart [10] We displayed high-saturationcolour by doubling two transparent charts [11]An integrating sphere was used to create a light sourcewith a uniform luminance distribution and the colourtemperature of the light source was 6810 K Figure 7shows the appearance of the integrating sphere and colourchart The colour samples from No 1 through No 24and from No 25 through No 43 were obtained usingone transmission chart and two doubling chartsrespectively

The tristimulus values of each colour patch were measuredusing a spectroradiometer (Topconrsquos SR-UL1) and were setas the reference values The colour differences (DE)between the reference values and the camera system valueswere calculated using a CIE LAB colour-differenceformula An LM was derived via a calculation whichminimises the average of the colour differences of thesecolour patches The following LM is used for colourcorrection of the capturing system

116 minus021 005minus002 112 minus010minus001 minus003 103

⎡⎣

⎤⎦

The average value and maximum of DE were 10 and 29(No 38) respectively Figure 8 shows the xy chromaticitiesof colour patches measured by the spectroradiometer

Figure 6 Example images of a moving object at frame frequencies of (a) 120 Hz and (b) 60 Hz

Figure 5 MTF characteristics of the camera

Figure 7 Integrating sphere and 24-patch transmissioncolour checker chart displaying high-saturation colour bydoubling two charts



(reference) and captured by the camera in the CIE 1931chromaticity diagram and standard RGB primaries Thenumbers beside the points are the patch numbers of thecolour checker

Studio equipment interfaceInterface requirements

The requirements of the interface used to connect toUHDTV devices are as follows

dagger The interface should support all UHDTV formatsspecified in Rec ITU-R BT2020 in the unified methodfor interface dissemination

dagger The interface must be compact so that it can beincorporated into UHDTV devices have low-poweroperation achieve signal transmission with a single cableand be easily connected and

dagger The interface should transmit signals at a transmissiondistance of approximately 50 m because the interfaceconnects devices in a studio or outside broadcast van

Interface design

The payload data-rate for the transmission of a full-specification SHV signal is approximately 144 Gbits It isdifficult to transmit a full-specification SHV signal via asingle coaxial cable with current technologies therefore weadopt an optical device

We adopted an optical multilink transmission systemwhich is composed of a multicore multimode fibre-opticcable and parallel fibre-optic transceiver because the aim ofthe interface is to transmit a UHDTV signal over a shortlength with low cost Furthermore we employed 10-Gbitsoptical transceivers which are practically available with

Figure 8 xy chromaticities of colour patches

Figure 9 Overview of the signal mapping of a full-specification SHV image into proposed 10-Gbits multilink interface

Table 3 Connector specifications

Number of fibres 24 (12 times 2)

Fibre type Multimode fibre

Connection loss Less than 075 dB

Insertionswithdrawals More than 5000 times

Equilibrium tensile loadingof connectors

250 N

Lock mechanism Bayonet-type lock



current technologies The target bit error rate is set to no lessthan 10212 which is defined by the existing standards using a10-Gbits signal [12]

Signal mapping

Figure 9 shows an overview of a full-specification SHV imagemapped into multilink 10-Gbits streams First each colourcomponent signal (R G B) is divided into four 4K sub-images with a sample structure of 3840 times 2160 pixelsVarious UHDTV signals (7680 times 4320 or 3840 times 2160pixels sampling structures of 444 422 and 420) can besimilarly handled using this method Next a 4K sub-imageis divided into basic-images Then the basic-stream isobtained by streaming a basic-image The basic-streams aremultiplexed and 8b10b encoding is performed 10-Gbitsdata are obtained by serialising the 8b10b encoding dataThe mapping method is capable of switching framesbecause the frame frequency of the original signal ismaintained namely the method supports a 120-Hz signal

On the basis of the aforementioned interface system aprototype interface which connects UHDTV devices by

using parallel fibre-optic transceivers and a multimodefibre-optic cable was developed Moreover the practicalityand feasibity of the system were verified [13]

Standardisation

This system was proposed to the Association of RadioIndustries and Businesses (ARIB) which is a Japanesestandardisation organisation and was standardised as ARIBSTD-B58 in March 2014 [14]

Because the intensity of the multiple-fibre push-on(MPO) ordinarily adopted in optical multilink transmissionis insufficient in terms of the strength for broadcasting usea new connector standard was determined We aimed toensure an operability and a shape similar to a BNCconnector Table 3 lists the specifications of the

Figure 10 Connector appearance

Figure 11 Appearance of the new interface optical cable(a) and a coaxial cable (b)

Table 4 Specifications of the wide-colour gamut laserprojector

Device 13 in Liquid Crystal on Silicon (LCOS)

Spatial resolutionpixels

Equivalent to 7680 times 4320


Light source 3000 ANSI lumen

Contrast 5000 1

Bit depth 12 bits

Light source RGB semiconductor laser (R 639 nmG 532 nm B 462 nm)

Table 5 ndash Primaries of the wide-colour gamut laserprojector

CIE x CIE y

Red primary 07157 02837

Green primary 01781 07846

Blue primary 01342 00455

Figure 12 ndashWide-colour gamut projector



standardised connector and its appearance in shown inFigure 10 We developed the full-specification SHVcamera system described in the previous section a videorouting switcher and a display which incorporates theinterface Further we also developed an optical cable forconnecting equipment Figure 11 shows the appearance ofthe optical cable compared to a coaxial cable where theoptical cable is similar in size to the coaxial cable

Monitoring equipmentWe have developed a wide-colour gamut SHV projectoroperating at 120 Hz by modifying a prefabricated projector(15) The light sources of the SHV projector were replacedwith red green and blue lasers corresponding to the wide-gamut colorimetry The wide-colour gamut projectorspecifications are listed in Table 4 and the projectorappearance is shown in Figure 12 The wavelengths of thelasers used as the light sources are slightly different fromthe wavelengths corresponding to the three primary coloursdefined by the standardised format however they wereadopted in consideration of the cost and feasibility of thedevices using present technology The chromaticity pointsof the three primary colours displayed by the projector weremeasured using a spectroradiometer The results are listedin Table 5 and it is confirmed that the projector achievedapproximately equal primaries to Rec ITU-R BT2020We also developed a waveform monitor which displays thexy chromaticity distribution of the wide-gamut RGBsignals (Figure 13)

Additionally we also developed an 85-inch liquid crystaldisplay (LCD) with 8K resolution operating at 120 Hzand incorporating the new interface described in theprevious section The colour gamut of the LCD isconsiderably wider than that of HDTV but needs to be

Figure 14 System diagram of full-specification SHV demonstration

Figure 15 Full-specification SHV demonstration at the NHK STRL Open House 2014

Figure 13 Waveform and colour gamut monitor



further widened to achieve that of Rec BT 2020 We plan toimprove the colour filters and the backlights of the LCD

Demonstration of full-specificationSHVWe used the aforementioned equipment to produce full-specification SHV content to evaluate the video qualityThe effectiveness of a full-specification SHV wasconfirmed in a demonstration at the NHK Science andTechnology Research Laboratories (NHK STRL) OpenHouse held in May 2014 [16] Figure 14 shows thedemonstration system The new interface is incorporated inall of the equipment (camera signal player routingswitcher and LCD) which all operates at a framefrequency of 120 Hz The improvements in the imagequality of an 8K resolution 120-Hz frame frequency andwide-colour gamut system were verified Figure 15 showsthe appearance of the demonstration

ConclusionsIn this paper we have described the equipment developed tocreate a full-specification SHV system the worldrsquos first full-specification camera system an interface and a display

The camera system can capture 8K resolution images at a 120-Hz frame frequency We developed an interface which cantransmit a full-specification SHV signal with a data-rate ofapproximately 144 Gbits via a single optical cable We planto devise a monitoring method for the optical signals anddiscuss the possibility of audio data insertion into an ancillarysignal domain We also developed monitoring devices an8K120Hz LCD 8K120Hz wide-colour gamut laserprojector and a waveform monitor which supports wide-gamutsignals We confirmed the capability of the full-specificationSHV system and the effectiveness of the 8K resolution120-Hz frame frequency and wide-colour gamut colorimetry

References

[1] Recommendation ITU-R BT2020 2012 ParameterValues for Ultra-High Definition Television Systems forProduction and International Programme Exchange

[2] SMPTE ST 2036-1 2013 Ultra High DefinitionTelevision ndash Image Parameter Values for Program Production

[3] ARIB STD-B56 (11) 2014 UHDTV System Parametersfor Programme Production March 2014

[4] NISHIDA Y MASAOKA K SUGAWARA M OHMURA M EMOTO MNAKASU E lsquoSuper Hi-Vision System Offering Enhanced

Sense of Presence and New Visual Experiencersquo The Bestof IET and IBC 2011 3 (2011) pp 5ndash10

[5] YAMASHITAT FUNATSU R YANAGI T MITANI K NOJIRI Y YOSHIDA TlsquoA Camera System Using Three 33-Megapixel CMOS ImageSensors for UHDTV2rsquo SMPTE Motion Imaging JournalNovemberDecember 2011 pp 24ndash30

[6] Recommendation ITU-R BT709-5 2002 ParameterValues for the HDTV Standards for Production andInternational Programme Exchange

[7] MASAOKA K SOENO T KUSAKABE Y YAMASHITA T NISHIDA YSUGAWARA M lsquoUHDTV System Colorimetry and Wide-Gamut Color Reproductionrsquo Asia-Pacific BroadcastingUnion TECHNICAL REVIEW January-March 2014pp 2ndash7

[8] ISO-12233 Photography - Electronic Still-PictureCameras Resolution Measurements

[9] BOREMAN GD lsquoModulation Transfer Function in OpticalandElectro-Optical Systemsrsquo SPIE PRESS 2000 pp 31ndash57

[10] TE188 Color Rendition Chart Image EngineeringFrechen Germany

[11] MASAOKA K NISHIDA Y SUGAWARA M KATO S SAITA AlsquoSpectral Sensitivity and Color Reproduction for SHVWide-Gamut Camerarsquo ITE Technical Report 2011 Vol 35No 45 pp 83ndash86

[12] SMPTE ST 435-2 2012 10 Gbs Serial DigitalInterface ndash Part2 10692 Gbs Stream ndash Basic StreamData Mapping

[13] SOENO T NISHIDA Y YAMASHITA T KUSAKABE Y FUNATSU RNAKAMURA T lsquoDevelopment of a Multilink 10 GbitsecMapping Method and Interface Device for 120 Framessec Ultrahigh-Definition Television Signalsrsquo SMPTE MotionImaging Journal MayJune 2014 pp 29ndash38

[14] ARIB STD-B58 (10) 2014 Interface for UHDTVProduction Systems March 2014

[15] SHIMAMOTO H KITAMURA K WATABE T OHTAKE H EGAMI NKUSAKABE Y NISHIDA Y KAWAHITO S KOSUGI T WATANABE T YANAGI TYOSHIDA T KIKUCHI H 2012 lsquo120 Hz Frame-Rate Super Hi-Vision Capture and Display Devicesrsquo SMPTE MotionImaging Journal March 2013 pp 55ndash61

[16] NHK Science and Technology Research LaboratoriesOpen House 2014 Full-Specification 8K Super Hi-VisionProduction System




ISSN 2041-5923

HEVC subjective video quality test resultsTK Tan1 M Mrak2 R Weerakkody2 N Ramzan3

V Baroncini4 GJ Sullivan5 J-R Ohm6 KD McCann7

1NTT DOCOMO Inc Japan2British Broadcasting Corporation UK3University of West of Scotland UK4Fondazione Ugo Bordoni Italy5Microsoft Corporation USA6RWTH Aachen University Germany7Zetacast UK

Abstract The primary goal for the development of the new High Efficiency Video Coding (HEVC) standard was toachieve a substantial improvement in compression efficiency relative to its predecessor H264AVC Howevermost comparisons to date had used simple objective measurements rather than a formal subjectiveevaluation Though objective tests are relatively easy to perform their results can only be indicative of theactual quality perceived by viewers

This paper describes the verification tests that were conducted by the Joint Collaborative Team on VideoCoding (JCT-VC) of ITU-T SG 16 WP 3 (VCEG) and ISOIEC JTC 1SC 29WG 11 (MPEG) Both objective andsubjective evaluations were conducted comparing the HEVC Main profile to the H264AVC High profile Thetests used sequences with resolutions ranging from 480p to Ultra High Definition (UHD) encoded at variousquality levels The results showed that the subjective performance of HEVC is actually better than would bepredicted from the objective test results In 86 of the test cases the subjective test results showed thatHEVC required half or less of the bit-rate of the H264AVC reference to achieve comparable quality Theaverage bit-rate savings for test sequences at UHD 1080p 720p and 480p resolutions were found to beapproximately 64 62 56 and 52 respectively

IntroductionWe are currently witnessing something that has become aonce-in-a-decade event in the world of video compressionthe emergence of a major new family of video compressionstandards

The mid-1990s saw the introduction of MPEG-2 thefirst compression standard to be widely adopted inbroadcasting applications H264AVC appeared in themid-2000s offering the same subjective quality atapproximately half the bit-rate Now a new standard HighEfficiency Video Coding (HEVC) has been developedthat promises a further factor of two improvement incompression efficiency for the mid-2010s As aconsequence HEVC can likely be used to broadcast Ultra

High Definition (UHD) video while using transmissionbit-rates that a decade ago were required for HighDefinition (HD) and two decades ago for StandardDefinition (SD) resolution

The HEVC standard has been jointly developed by thesame two standardization organizations whose previouscollaboration resulted in both MPEG-2 and H264AVCthe ISOIEC Moving Picture Experts Group (MPEG)and the ITU-T Video Coding Experts Group (VCEG)

The initial edition of the HEVC standard was completedin January 2013 and it is published by ISOIEC as ISOIEC23008-2 (MPEG-H Part 2) and by ITU-T asRecommendation H265 [1] This first version supportsapplications that require single-layer 420 sampled video



with 8 or 10-bit precision A second edition which extendsthe standard to support contribution applications with toolsthat enable 422 and 444 sampled video as well as 10 12or 16-bit precision was completed in April 2014 Furtheramendments are currently being developed to extend thestandard by efficient scalability and stereomultiviewcompression and to allow representation of 3D (video plusdepth) content

This paper presents results obtained during the recentJCT-VC verification testing of HEVC The purpose ofthis test was to verify that the key objective of HEVC hadbeen achieved providing a substantial improvement incompression efficiency (eg by a factor of two) relative toits predecessor ITU-T Rec H264 | ISOIEC 14496-10Advanced Video Coding (AVC) [2] The test sequencesused for the verification testing were deliberately chosen tobe different from those that had been used during thedevelopment of the standard to avoid any possible bias thatthe standard could have towards those sequences

This paper is organized as follows The test methodsmaterial and laboratory set-up are described in the sectionof verification test methodology The subjective test resultsand analysis are described in the section on verificationtest results Finally the conclusions are drawn in the lastsection

Verification test methodologyTest methods

The subjective test method adopted for this evaluation isDegradation Category Rating (DCR) [3] This test methodis commonly adopted when the material to be evaluatedshows a range of visual quality that distributes well acrossall quality scales This method was used under the schemaof evaluation of the quality (and not of the impairment)for this reason a quality rating scale made of 11 levels wasadopted ranging from ldquo0rdquo (lowest quality) to ldquo10rdquo (highestquality)

The structure of the Basic Test Cell (BTC) of DCR isshown in Figure 1 The structure comprises twoconsecutive presentations of the video clip under test Firstthe original version of the video clip is displayed followedby the coded version of the video clip Then a messageasking the viewers to vote is displayed for 5 seconds Amid-grey screen displaying for one second precedes thepresentation of both video clips All the video materialused for these tests consists of video clips of 10 secondsduration

Test material

The test material for the verification test was encodedusing HM121 (the reference software of the HEVCcodec) and JM185 (the reference software of theAVC codec)

Four picture resolutions were tested UHD-1 (2160p)1080p 720p and 480p Each resolution was represented by5 test sequences giving a total of 20 test sequences

For each test sequence 4 test points with different bit-rateswere chosen to be encoded using the HM121 and JM185software The four JM185 bit-rates were chosen such thatthe quality levels spanned the entire range of the meanopinion score (MOS) scale The HM121 test points withindices i frac14 0 1 2 3 were then selected such that the bit-rates RHEVC(i) were approximately half the bit-rates of thecorresponding JM185 reference points RAVC(i) ieRHEVC(i) asymp 1

2 RAVC(i)

Further detail including the coding conditions used thequantization parameter (QP) selection and encoded bit-rates can be found in Annex B of [3]

Laboratory Setup

For subjective assessment the laboratories were set upaccording to [3] except for the selection of the display andthe video play-out server

The TVLogic LUM-560W 56rdquo professional LCDmonitor was used for UHD sequences the Pioneer 50rdquoPDP-5000EX monitor was used for 720p Both monitorswere set to their native resolutions of 3840 times 2160 and1920 times 1080 pixels respectively

For the 1080p and 480p video clips the monitor used wasthe HP 30rdquo ZR30W set at its native resolution of1920 times 1080 pixels Two simultaneous displays were usedwith one viewer per screen

Play-out of the 720p and 480p video clips was at the nativeresolution of the monitors (1920 times 1080) using the centralarea of the screen the remaining part of the screen is set toa mid-grey level (128 in 0ndash255 range)

The viewing distance varied according to the physicaldimensions and the native resolutions of the monitors thisled to the viewing distance varying from 15H to 3Hwhere H is equal to the height of the screen

Verification test resultsSubjective test results

In the analysis of the results the mean opinion score (MOS)values and the associated confidence interval (CI) werecomputed for each test point From the raw data theFigure 1 DCR BTC



reliability of each viewer was calculated by computing thecorrelation index between each score provided by a viewerto the general MOS value computed for that test point Acorrelation index greater than or equal to 075 wasconsidered as valid for the acceptance of the viewer

The MOS values together with the CI were then plottedagainst the bit-rate for each sequence tested Figures 2through 5 show examples of the plots for each of theresolutions tested The complete data set of the twentysequences can be found in [3]

Figure 2 Example of UHD sequence

Figure 3 Example of 1080p sequence




Analysis

All the plots in Figures 2 through 5 show that the rate-distortion (RD) curves of the HEVC test points are locatedsubstantially to the left of the RD-curves of the AVCreference points This shows that HEVC is achieving asubstantial bit-rate reduction relative to AVC

An example of the analysis of the test points is shown inFigure 6 All the test points were classified into thefollowing cases where the HEVC test points PHEVC(i)have the same subjective quality (overlapping MOSconfidence intervals (CI)) as the AVC reference pointsPAVC(i) where the bit-rates of AVC are

(a) greater than double the bit-rates of HEVC (CIAVC(m)overlaps with CIHEVC(n) and m n)(b) equal to double the bit-rates of HEVC (CIAVC(m)overlaps with CIHEVC(n) and m frac14 n)(c) less than double the bit-rates of HEVC (CIAVC(m)overlaps with CIHEVC(n) and m n) and CI(i)corresponds to the confidence interval of R(i) and both areassociated to test point P (i)

45 of the HEVC test points achieved approximately 50better coding efficiency than AVC Another 41 of theHEVC test points achieved more than 50 better codingefficiency than AVC In contrast only 14 of the HEVCtest points achieved a coding efficiency of less than 50better than AVC

Even though the inspection of the RD-curves and theoverlapping MOS CI gave high confidence that HEVCwas achieving a bit-rate reduction over AVC that was verylikely above 50 the test data points were not sufficientlydense to quantify precisely the amount of the bit-ratereduction In order to estimate the coding efficiencyimprovement achieved the approach described in the nextsection was employed

MOS BD-rate

In this section the average bit-rate savings of HEVCcompared to AVC for each sequence are computed fromthe MOS scores in the same manner as in [5][6] tofurther quantify the bit-rate savings achieved TheBjoslashntegaard Delta (BD) rate measure described in [7][8]

Figure 6 Examples of test points with overlapping CI




was used with the MOS scores taking the place of the PeakSignal-to-Noise Ratio (PSNR) and a piecewise cubicinterpolation instead of cubic spline interpolation

Average bit-rate savings were calculated where the sameMOS scores for both HEVC and AVC could beinterpolated from subjective test results as shown in

Table 1 Average bit-rate savings

Resolution Sequence MOS BD-rate

UHD-1 (2160p) BT709Birthday 275

Book 266

HomelessSleeping lowast

Manege 256

Traffic 258

1080p JohnnyLobby (LD) 270

Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268

720p ThreePeople (LD) 248

BT709Parakeets 266

QuarterBackSneak 258

SVT01a 273

SVT04a 236

480p Cubicle (LD) 245

Anemone 242

BT709BirthdayFlash 249

Ducks 272

WheelAndCalender lowast

Average 2587

Figure 7 Average bit-rate savings (measured by BD-Rate) of HEVC compared to AVC



Figures 2 through 5 The interval over which the BD-rate isaveraged is shown as solid lines in each of these plots Table 1shows the average bit-rate savings calculated by this measurefor the 20 sequences

Bit-rate savings were not calculated for the sequencesHomelessSleeping and WheelAndCalender since they didnot satisfy the criteria for an accurate calculation MOSvalues exhibiting a smooth curve with an averaging intervalinterpolated from at least three MOS scores withmonotonically increasing bit-rate

The average bit-rate saving calculated from these resultsconfirmed that the HEVC Main profile achieved the samesubjective quality as AVC High profile while requiringapproximately 59 fewer bits It was noticeable that higherresolution sequences tended to achieve slightly greatersavings than lower resolution sequences

Figure 7 shows the average bit-rate savings for testsequences with UHD-1 1080p 720p and 480presolutions which are estimated at approximately 6462 56 and 52 respectively The bit-rates shown inthis illustration correspond to the average of the highestbit-rate points over all sequences at each resolution

Table 3 of [3] shows that the BD-Rate savings remainconsistently high when only the results of test points withquality suitable for broadcasting (MOS greater than or equalto 7 corresponding to ldquogoodrdquo or ldquoexcellentrdquo) are considered

ConclusionsAnalysis of the subjective test results show that HEVC at halfor less than half the bit-rate of the AVC reference achievedcomparable quality in 86 of the test cases

By applying a MOS BD-rate measurement on the resultsof the subjective test it was found that the HEVC Mainprofile achieves the same subjective quality as AVC Highprofile while requiring on average approximately 59 fewerbits

It can therefore be concluded that the HEVC standard isable to deliver the same subjective quality as AVC while onaverage only requiring half or even less than half of the bit-rate used by AVC The initial objective of the HEVCdevelopment of substantial improvement in compressioncompared to previous state of the art has therefore beensuccessfully achieved

AcknowledgmentThe authors thank the contributors of the video test sequencematerial for permitting its use for the tests and thispublication and their collaborators in the JCT-VC fortheir assistance in developing the test plan and analyzing itsresults

References

[1] ITU-T and ISOIEC JTC1 ldquoHigh Efficiency Video CodingITU-T Rec H265 and ISOIEC 23008-2rdquo April 2013

[2] ITU-T and ISOIEC JTC1 ldquoAdvanced Video Coding forGeneric Audio-Visual Services ITU-T Rec H264 and ISOIEC 14496-10 (AVC)rdquo March 2005

[3] TAN TK ET AL lsquoHEVC verification test reportrsquo JointCollaborative Team on Video Coding (JCT-VC) of ITU-T SG16 WP 3 and ISOIEC JTC 1SC 29WG 11 JCTVC Q1011(ISOIEC JTC1SC29WG11 N14420) 17th MeetingValencia ES 27 March ndash 4 April 2014

[4] International Telecommunication UnionStandardization Sector Recommendation ITU T P910ldquoSubjective video quality assessment methods formultimedia applicationsrdquo

[5] TAN TK FUJIBAYASHI A SUZUKI Y TAKIUE J lsquoObjective andSubjective Evaluation of HM50rsquo document JCTVC-H0116Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG 16 WP 3 and ISOIEC JTC 1SC 29WG 11 San JoseCA February 2012

[6] OHM J-R ET AL lsquoComparison of the Coding Efficiency ofVideo Coding Standards ndash Including High Efficiency VideoCoding (HEVC)rsquo invited paper to IEEE CSVT Special Issueon Emerging Research and Standards in Next GenerationVideo Coding (HEVC) December 2012

[7] BJOslashNTEGAARD G lsquoCalculation of Average PSNRDifferences Between RD Curvesrsquo ITU T SG16Q6 13th

VCEG Meeting Austin Texas USA April 2001 DocVCEG-M33

[8] BJOslashNTEGAARD G lsquoImprovements of the BD-PSNR modelrsquoITU-T SG16Q6 35th VCEG Meeting Berlin Germany July2008 DocVCEG-AI11



ISSN 2041-5923Introduction to Electronics Letters

In 2014 Electronics Letters is celebrating its 50th year of publication Launched in 1965 just two years before the first IBC overthe last five decades Electronics Letters has published over 43000 papers and seen its scope evolve to reflect the amazing changesand advances in electronics since the 1960s

Electronics Letters1 is a uniquely multidisciplinary rapid publication journal with a short paper format that allows researchers toquickly disseminate their work to a wide international audience Electronics Letters broad scope involves virtually all aspects ofelectrical and electronic technology from the materials used to create circuits through devices and systems to the softwareused in a wide range of applications The fields of research covered are relevant to many aspects of multimedia broadcastingincluding fundamental telecommunication technologies and video and image processing

Each year the Electronics Letters editorial team and the executive committee of the IET Multimedia CommunicationsCommunity come together to select a small number of papers from the relevant content of Electronics Letters to appear inthis publication

This year all three of the chosen papers were also highlighted in the free magazine style news section included as part of eachissue of Electronics Letters The news section includes articles based on some of the best papers in each issue providing morebackground and insight into the work reported in the papers The news articles associated with all three papers are also includedin this 2014 volume of The Best of IET and IBC

Respectively addressing terahertz communications issues encountered in evolving current optical networks for new servicesand developing components for increasingly multi-standard wireless devices the papers selected give an idea of how researchpublished in Electronics Letters will impact both the immediate and long term future of the broadcast media industry

We hope you will enjoy reading these papers and features as examples of our content and if you like what you read all ourfeature articles are available for free via our web pages1

The Electronics Letters editorial team

1wwwtheietorgeletters

65 The Best of IET and IBC 2014 Vol 6 p 65


Selected content from Electronic Letters

ISSN 2041-5923

Coherent THz communication at 200 GHzusing a frequency comb UTC-PD andelectronic detectionG Ducournau1 Y Yoshimizu2 S Hisatake2 F Pavanello1

E Peytavit1 M Zaknoune1 T Nagatsuma2 J-F Lampin1

1Institut drsquoElectronique de Microelectronique et de Nanotechnologie (IEMN) UMR CNRS 8520 Universite Lille 1 AvenuePoincare BP 60069 59652 Villeneuve drsquoAscq France2Graduate School of Engineering Science Osaka University 1-3 Machikaneyama Toyonaka 560-8531 JapanE-mail nagatumaeeesosaka-uacjp

Abstract A coherent terahertz (THz) link at 200 GHz with a variable data rate up to 11 Gbits featuring a veryhigh sensitivity at the receiver is investigated The system uses a quasi-optic unitravelling carrier photodiode(UTC-PD) emitter and an electronic receiver The coherent link relies on an optical frequency comb generatorat the emission to produce an optical beat note with 200 GHz separation phase-locked with the receiver Biterror ratio testing has been carried out using an indoor link configuration and error-free operation isobtained up to 10 Gbits with a received power 2 mW

IntroductionTo serve new services such as video on mobile terminals thedemand for bandwidth is growing every year and new waysfor wireless data transmission are being investigatedTerahertz (THz) communications are thus very promising[1] as their open a huge space for new high data-rateservices THz communications have been intensivelystudied over the past few years and systems are nowemerging based on photonic andor electronic technologiesAmong these different systems coherent THz links pavethe way for high performances in terms of data rate andsensitivity at receiver circuits For example using photonicsat the emission Nagatsuma et al [2] have achieved real-time coherent 100 GHz up to 11 Gbits and Li et al [3]reported 200 Gbits in the W-band using off-linedemodulation and signal processing Above the W-bandKoenig et al [4] reported very recently up to 100 Gbitsfor a 20 m link at 2375 GHz also using the off-linedetection At the electronic side (both at emission andreception) Antes et al [5] achieved 30 Gbits at 240 GHzusing off-line processing Future real-time THz coherentcommunication systems will require a locking between the

transmitter (Tx) and the receiver (Rx) to supress anyresidual frequency drift between Tx and Rx downconvertedinto random amplitude modulation at the receiving mixerFor this reason off-line detection can be used to overcomethe locking problem in order to highlight the intrinsiccharacteristics of the THz link At the photonic side a veryconvenient way to create the dual-tone optical signal is touse optical frequency combs where the combs remainlocked on a microwave reference One of the key issues torealise coherent communication by photonics-based THzgeneration is the phase locking In the proposed systemthe phases of two optical sidebands extracted with anarrayed waveguide grating (AWG) have been locked to theoptical frequency comb

Experimental setupThe data link is described in Fig 1

Transmitter descriptionThe system is first composed of the optical frequency combgenerator (OFCG) described in Fig 2 The phase of a



single frequency continuous-wave (CW) laser is modulatedwith cascaded optical phase modulators to generate theOFC The modulation frequency was 166 GHz Dual-optical carriers with 200 GHz frequency separation areextracted with an AWG filter The phases of the twooptical carriers fluctuate independently due to temperaturefluctuations and acoustic noise in the optical fibre cablesbetween the AWG filter and the optical coupler Thecoherence of the THz wave to be generated as a carrier forthe wireless link is dependent on the stability of the phasedifferences between the optical carriers In our system thephases of the two carriers are locked to that of the OFC togenerate a coherent THz wave the spectrum of which isgiven in Fig 3 Details of the phase stabilisation system aredescribed in [6 7]

The dual-optical signal is then modulated by a Mach-Zehnder amplitude modulator with a variable data rate upto 11 Gbits amplified with an erbium-doped fibreamplifier and feeds a quasi-optic unitravelling carrierphotodiode (UTC-PD) module as shown in Fig 4 ThisUTC-PD has a 02 AW responsivity and is integratedwith a log-periodic antenna First the quasi-optic UTC-PD module was power calibrated in free space using an

Erickson PM4 Using a 1 mA photocurrent the modulatedTHz signal emitted power was found to be around 3 mW(225 dBm) at 200 GHz and 15 mW with a 07 mAphotocurrent The THz path was composed of twopolymer lenses This THz path was preliminarily testedusing a quasi-optical vector network analyser (VNA)leading to THz losses of 3 dB at 200 GHz frequencyHowever it is worth mentioning that this value justestimates the losses as the first lens was fed by a waveguidehorn antenna during the VNA experiment and was fed bythe integrated silicon lens during the data transmissionexperiments

Receiver descriptionThe receiver is first composed of a WR51 waveguidecorrugated horn (Fig 4) The detected signal feeds a sub-harmonic mixer (SHM) pumped with a local oscillator at100 GHz This SHM exhibits 8 dB conversion losses at200 GHz The intermediate frequency (IF) signal is thenamplified with a wideband amplifier (33 dB gain) Whenthe optical beat note is controlled with the phasestabilisation system the 200 GHz carrier is phase-locked tothe reference synthesiser At the receiver side the mixingproduced is centred at DC as long as the carrier frequencyremains in phase with the local oscillator Under theseconditions the baseband data are recovered and a limiting

Figure 1 Experimental setup of 200 GHz coherent data link

CW continuous wave Df fixed electrical phase shifter

Figure 2 OFCG details

Figure 3 Optical spectra obtained at OFCG output

Figure 4 Views of quasi-optic UTC-PD module andelectronic receiver working at 200 GHz

a Quasi-optic UTC-PD moduleb Electronic receiver working at 200 GHz



amplifier is used before eye diagrams display or bit error ratio(BER) measurements

ResultsIn the experiment the coherent operation was obtained usingthe same reference to feed the comb generator and thereceiver circuits This enables one to address easily the real-time capability of the THz link without off-line dataprocessing Indeed we succeed to realise true BER testingusing the combination of the MP1775A tester (Anritsu)for psuedorandom binary sequence (PRBS) generation andthe MP1776A (Anritsu) for BER measurement with215 2 1 bit sequences up to 11 Gbits Up to 10 Gbitsdata rate the slope of the BER curves is similar for thedifferent tested data rates as shown in Fig 5 Moreover an3 dB penalty was obtained when increasing the data ratefrom 5 to 10 Gbits Indeed an increase of the data rate bya factor of 2 while keeping the signal constant shoulddecrease the signal-to-noise ratio by 3 dB The error-freeoperation (defined here for BER 10211) is obtained at5 Gbits for 07 mA in the UTC-PD corresponding to a15 mW power at the UTC-PD output This shouldcorrespond to 075 mW 230 dBm) received power For10 Gbits the error-free operation was obtained with 1 mAphotocurrent in the UTC-PD corresponding to 228dBm received power

The BER curve obtained for the 11 Gbits data ratepresented a reduced slope and the achieved BERperformance was limited to few 1026 We assume that thiscomes from the limited bandwidth available at theheterodyne receiver To verify this assumption a frequencyresponse of the receiving mixer was carried out In thisexperiment the mixer was tested under waveguideconditions using a continuous- wave source coming from aVNA and the frequency was tuned between 180 and220 GHz Fig 6 presents the results From this curve a20 GHz bandwidth +10 GHz around the carrier

frequency) enables an up to 10 Gbits error-free operationFor higher data rates a very large amount of amplitudedistortion affects the receiver which is highlighted inFig 6 by an increase of the conversion losses (decreasingIF response) For that reason the eye diagrams beyond 10Gbits were degraded and the BER performance waslimited In our last experiment an uncompressed high-definition television data stream (15 Gbits) was used totest the system with real applicative signals and the stableoperation (error-free video transmission) was obtained upto 06 mA photocurrent corresponding to 500 nWreceived power

ConclusionA coherent transmission system was demonstrated at200 GHz using a photonics-based emission and featuringa very low power requirement at the reception Error-freeoperation was obtained up to 10 Gbits data rates with alimitation due to the receiver bandwidth not by the combor the UTC-PD transmitter Next steps will concern highdistance transmission using powerful photomixers at theemission Moreover in a future outdoor THz coherentscheme application locking on the global positioningsystem (GPS) will lead to a common reference between Txand Rx avoiding the need for the common referenceused here

AcknowledgmentsThis work has been partially supported by the AgenceNationale de la Recherche and Japanese Science andTechnology Agency within the framework of the lsquoWITHrsquoANRJST Franco-Japanese project and was partlysupported by the Ministry of Higher Education andResearch Nord-Pas de Calais Regional Council andFEDER through the Contrat de Projets Etat RegionlsquoCampus Intelligente Ambiantersquo (CPER-CIA) 2007ndash2013

Figure 5 BER characteristics using 11 Gbits limitingamplifier

Inset Eye diagrams for highest photocurrents tested

Figure 6 Frequency response of heterodyne receiverincluding IF wideband amplifier



References

[1] FEDERICI J MOELLER L lsquoReview of terahertz andsubterahertz wireless communicationsrsquo J Appl Phys2010 107 p 111101

[2] NAGATSUMA T HORIGUCHI S MINAMIKATA Y YOSHIMIZU YHISATAKE S KUWANO S YOSHIMOTO N TERADA J TAKAHASHI HlsquoTerahertz wireless communications based on photonicstechnologiesrsquo Opt Express 2013 21 (20) p 23736

[3] LI X YU J ZHANG J DONG Z LI F CHI N lsquoA 400G opticalwireless integration delivery systemrsquo Opt Express 201321 (16) p 187894

[4] KOENIG S LOPEZ-DIAZ D ANTES J BOES F HENNEBERGER RLEUTHER A TESSMANN A SCHMOGROW R HILLERKUSS D PALMER RZWICK T KOOS C FREUDE W AMBACHER O LEUTHOLD J KALLFASS I

lsquoWireless sub-THz communication system with high dataratersquo Nat Photonics doi 101038NPHOTON2013275

[5] ANTES J KOENIG S LOPEZ-DIAZ D BOES F TESSMANN AHENNEBERGER R AMBACHER O ZWICK T KALLFASS I lsquoTransmissionof an 8-PSK modulated 30 Gbits signal using an MMIC-based 240 GHz wireless linkrsquo IMS Conf Seattle WA USAJune 2013

[6] YOSHIMIZU Y HISATAKE S KUWANO S TERADA J YOSHIMOTO NNAGATSUMA T lsquoWireless transmission using coherentterahertz wave with phase stabilizationrsquo IEICE ElectronExpress 2013 10 (18) p 20130578

[7] YOSHIMIZU Y HISATAKE S KUWANO S TERADA J YOSHIMOTO NNAGATSUMA T lsquoGeneration of coherent sub-terahertz carrierwith phase stabilization for wireless communicationsrsquoJ Commun Netw 2013 15 (6) pp 569ndash575




ISSN 2041-5923

Minimising nonlinear Raman crosstalk infuture network overlays on legacy passiveoptical networksD PiehlerNeoPhotonics 2911 Zanker Road San Jose CA 95134 USAE-mail davidpiehlerneophotonicscom

Abstract There is a desire to overlay future optical access networks onto legacy passive optical networks (PONs)to provide increasingly advanced services by filling empty wavelength bands of legacy gigabit rate PONsNonlinear Raman crosstalk from new wavelengths onto legacy RF video services (1550 ndash 1560 nm band) andonto the legacy digital downstream at 1490 nm however can limit the number and the launch power of newwavelengths As an example straightforward physical-layer adjustments at the optical line terminal thatincrease the number of new 10 Gbits channels launched in the 1575 ndash 1580 nm band by a factor of 16without increasing the Raman penalty on the video signal are illustrated A physical-layer (RF) filter modifiesthe on ndashoff-keyed signal feeding each 10 Gbits transmitter suppressing the RF Raman crosstalk on the videosignal by 9 dB while incurring a power penalty on each 10 Gbits link of 05 (20) dB with (without) forwarderror correction The previous Raman mitigation work used non-standard line-coding to shape the 10 Gbitselectrical spectrum In addition polarisation-interleaving of new wavelengths lowers the worst-case RF DCcrosstalk by 3 dB in fibres and it limits and stabilises DC crosstalk in low polarisation mode dispersion fibre links

IntroductionPassive optical networks (PONs) with broadband opticalpower splitters assign discrete wavelength bands to variousdownstream and upstream services In principleunallocated wavelength bands can be utilised for overlayingfuture (higher-bandwidth) services on the legacy fibre plantwithout disrupting the legacy services It is clear that theterminal equipment of both the legacy and future networksmust provide a certain level of optical isolation betweensignals to limit any service degradation due to lsquolinearrsquooptical crosstalk This Letter outlines two steps that willlimit the impact of lsquononlinearrsquo optical crosstalk due to thenonlinear Raman effect when launching multiplewavelengths carrying new services into a legacy PON

As an example this Letter considers the nonlinear Ramanlimitations associated with launching multiple 10 Gbitsnon-return-to-zero (NRZ) modulated signals in the 1575ndash1580 nm band into a legacy gigabit rate PON co-

propagating with a 1550ndash1560 nm RF video overlay and a125 or 25 Gbits 1490 nm digital (NRZ) downstreamsignal

The nonlinear Raman effect mediates the transfer ofoptical power from shorter to longer wavelengths (DCcrosstalk) while the RF crosstalk flows in both directionsThe effect is robust requiring no phase matching and itcan facilitate network-failing interference betweenwavelengths separated by up to 220 nm [1] The mostwell-known impact of RF crosstalk in PONs is degradationof the carrier-to-noise ratio of a broadcast 1555 nm RFvideo signal (17 dBm launch power) from a relativelyweak (3 dBm launch power) co-propagating 1490 nmdigital signal [2] The less well-known DC Ramancrosstalk can cost the downstream 1490 nm signal one ormore dB in optical power by the time it reaches asubscriber [3] Since PON standards assume equal lossbudgets for upstream and downstream ignoringwavelength-dependent fibre loss the 1490 nm downstream


signal has an 010 dBkm lsquoeffectiversquo fibre gain compared tothe 1310 nm upstream This more than compensates for anyactual DC Raman loss

The number and the total optical power of 10 Gbits signalsare limited by both the DC crosstalk (Raman loss) whichattenuates the legacy 1490 nm signal and by the RFcrosstalk onto legacy RF video channels This Letter outlinestwo tools which when used together allow 16 10 Gbitswavelengths to be launched compared to one without usingthe tools with a minimal impact on the 10 Gbits links

The first tool the PHY-filter modifies the RF powerspectral density of each 10 Gbits NRZ signals in the 50ndash300 MHz region where the RF Raman crosstalk is mostsevere Earlier work by Colella et al shows that spectralshaping of the interfering electrical spectrum bymodification of its bit-patterns reduces RF Raman crosstalk[4] In this Letter the spectrum is shaped at the physical-layer without changing the bit-pattern meaning that theexisting bit-encoding and decoding standards and siliconcan be used at the optical line terminal (OLT) and at thesubscriber terminals The second tool involves minimisingpolarisation-dependent Raman crosstalk by polarisation-interleaving the new wavelengths

RF PHY-filterThe RF crosstalk on the video channel is proportional to afrequency-dependent walk-off parameter which depends onthe fibre interaction length attenuation chromaticdispersion and separation (Dl) between the videowavelength and the interfering wavelength(s) [1 2] Sincethe optical power drops precipitously after the opticalsplitter the interaction length is that of the optical trunkline (OTL) between the OLT and the splitter

As illustrated in Fig 1 the walk-off parameter is generallyhighest at lower frequencies Thus nonlinear Raman

crosstalk impacts the low-frequency RF video channelswhich begin at 50 MHz The intention of the PHY-filter is to lower the power spectral density of theinterfererrsquos electrical spectrum in this region In principlethis filter can be an RF filter placed between thetransmitterrsquos electronic driver and its laser or opticalmodulator From Fig 1 a triangular filter with a maximumattenuation at 50 MHz followed by a +005 dBMHzramp will accommodate a wide variety of deploymentscenarios

Fig 2 illustrates how various PHY-filter shapes modify thesinc-shaped RF spectrum of a 10 Gbits NRZ signal As longas each of the new 10 Gbits transmitters are fed byuncorrelated bit streams Raman crosstalk addsincoherently meaning that a 9 dB drop in crosstalk fromeach transmitter allows their number to increase by a factorof 8 while keeping the total RF power transferred to thevideo signal constant

Impact of PHY-filter on 10 GbitslinkFor baseband systems in general flat passbands and well-shaped high-frequency cutoffs are design objectives andirregularities in the frequency response are rarely studied[5 6] The impact from the triangular RF filters of Fig 2on a 10 Gbits 218ndash1 pseudorandom bit stream (PRBS)was numerically simulated by an application in thefrequency-domain onto the Fourier transform of the NRZPRBS An inverse Fourier transform produced an outputtime-domain waveform The PHY-filter has two maineffects on the bit stream The primary impact is that afiltered random bit stream contains a (Gaussian) noisewaveform Sfilter(t) with a power spectral density in the samefrequency band as the PHY-filter

As a second-order effect the post-filter waveform havinglost energy exhibits overall amplitude compression This

Figure 1 Calculated walk-off factor [1 2] against RFfrequency for OTL distances from 1 to 20 km and Dl frac14 15to 30 nm

Solid black line is maximum value for all scenarios (Assumptionsfibre dispersion frac14 18 ps(nm km) attenuation frac14 02 dBkm)

Figure 2 Power spectral density of PHY-filtered 10-GbitsNRZ signal for filters of depth 3 6 9 and 12 dB andpositive slope of 005 dBMHz



Letter defines the amplitude of the post-filtered signal as d frac14kPonelndash kPzerol (PonendashPzero 1 for the pre-filter bit stream)The measured values of d and the standard deviation s ofthe noise amplitude for the waveforms generated by thePHY-filters on the NRZ PRBS are listed in Table 1

With this information standard techniques are used tocalculate the optical power penalty due to the PHY-filterAssumptions on the optical transmitter and the receiverare laser transmitter RIN ndash140 dBHz extinction ratio10 dB receiver sensitivity 09 AW thermal noise 28 pAHz

12 avalanche photodiode (APD) multiplication factor

15 APD ionisation ratio 07 and electrical bandwidth67 GHz

If forwarded error correction (FEC) is (is not) used toimprove a 1023 bit error ratio (BER) to 10212 9 dB ofnonlinear Raman crosstalk cancelation will induce a 025(149) dB optical power penalty Thus 9 dB of (RF)Raman crosstalk reduction is possible with a modest powerpenalty on the interfering 10 Gbits signals especially ifFEC is used

Low-frequency filtering is known to impact the non-random bit streams [6] The example PHY-filters avoidedthe 50 MHz region to minimise the impact of longconsecutive-bit strings The simulated 9 dB triangle filtergenerated the Gaussian noise waveforms when up to 30-digit sequences of all ones or zeros were randomly added tothe PRBS

Although the PHY-filter may be realised as an actual RFfilter digital signal processing and an appropriate digital-to-analogue converter (DAC) can convert the PRBS intothe post-filter waveform A more compact and cost-effective realisation computes the PHY-filterrsquos noisewaveform Sfilter (t) in real-time from the PRBS createsthis analogue waveform with a 100 to 300 MHz DACand applies it to the unfiltered 10 Gbits NRZ bit streamas a (low-frequency) bias

Polarisation-dependent RF RamancrosstalkNonlinear Raman crosstalk is highly polarisation-dependentinvolving only parallel components of interacting lightwavesPolarisation mode dispersion (PMD) scrambles relativepolarisation states of signals at optical frequencies v1 andv2 as they co-propagate along a fibre of length z andPMD Dp according to the autocorrelation ks(z v1)

s(z v2)l = exp minusD2pz (Dv)2

3[ ]

[7] In a low-PMD fibre

link (Dp frac14 002 pskm12) all signals in the 1575ndash1580 nm

band will remain 95 polarisation-correlated for the first27 km If the 10 Gbits signals are launched with allpolarisations aligned the polarisation-dependent Ramancrosstalk (PDRXT frac14 RXTmaxndashRXTmin [in dB] over allpolarisations) can be calculated according to [8] Fig 3illustrates the calculated statistical distributions of PDRXTfor RF crosstalk from the 1577 nm band onto a videosignal at the (worst-case) wavelength of 1560 nm forvarious OTL lengths using a Raman gain coefficient gR of022(Wkm) and a total launch power (in the 1577 nmband) of +20 dBm

At all OTL lengths the expected values of PDRXTRXTav are nearly 2 indicating a lsquoworst-casersquo polarisationRF Raman crosstalk that is twice the average crosstalk(RXTav frac14 [RXTmax + RXTmin]2 [in dB]) If howeverthe signals in the 1577 nm band are polarisation-interleaved and launched with alternating polarisationstates the expected PDRXTRXTav would be nearly zeroand the worst-case RF crosstalk would be the averageRaman crosstalk Thus polarisation-interleaving enablesthe number of new 10 Gbits in the 1577 nm bandwavelengths to double (to 16) without changing thelsquoworst-casersquo RF crosstalk

Table 1 Calculated optical power penalties at 1023 and 10212 BER for 218 ndash1 PRBS with various triangular filter depths

Filter depth (dB) RF pass () s d Penalty 1023 BER (dB) Penalty 10212 BER (dB)

3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378

Figure 3 Probability density for polarisation-dependentRaman crosstalk when 1577 nm signals are launchedwith aligned polarisation states



Polarisation-dependent DC RamancrosstalkA simple calculation indicates that +20 dBm net powerlaunched in the 1577 nm band will cause a (polarisation-averaged) Raman loss of 18 dB in a co-propagating 1490 nmsignal at the end of a 20 km OTL ( gR frac14 061(Wkm))Fig 3 shows that with polarisation-aligned 1577 nmband signals the lsquoexpectedrsquo polarisation-dependent loss is07 equivalent to 24 dB loss at the lsquoworst-casersquopolarisation There is a small probability (4) that Ramanloss can be 3 dB and fluctuate over a 23 dB range aspolarisation states evolve over time If the 1577 nm bandsignals are polarisation-interleaved Raman loss is constantat 18 dB which can be compensated by the 01 dBkmlsquoeffectiversquo fibre gain mentioned in the Introduction SectionThe DC Raman crosstalk limits the net 1577 nm bandlaunch power to about +20 dBm For low-PMD linkspolarisation-interleaving may be required to minimise andstabilise the DC Raman crosstalk

ConclusionThe described PHY-layer techniques applied at the PONOLT can mitigate RF and DC Raman crosstalk when newwavelengths overlay legacy PONs Polarisation-interleavingmay be necessary to manage nonlinear Raman crosstalk inlow-PMD fibre plants

References

[1] PIEHLER D MEYRUEIX P RISHI G CHEN LP COPPINGER FlsquoNonlinear Raman crosstalk in a 125-Mbs CWDM overlayon a 1310-nm video access networkrsquo Optical FiberCommunications Conf Los Angeles CA USA February2004 paper FE8

[2] COPPINGER F CHEN LP PIEHLER D lsquoNonlinear Ramancross-talk in a video overlay passive opticalnetworkrsquo Optical Fiber Communications Conf Atlanta CAUSA March 2003 paper TuR5 doi 101109OFC20031247660

[3] AGATA A MURAKAMI A IMAI K HORIUCHI Y EDAGAWA NlsquoInfluence of stimulated Raman scattering in video overlayEPON systems on downstream data signalrsquo 31st EuropeanConf Optical Communications Amsterdam TheNetherlands September 2005 paper We457 doi101049cp200504165

[4] COLELLA B EFFENBERGER FJ SHIMER C TIAN F lsquoRamancrosstalk control in passive optical networksrsquo OpticalFiber Communications Conf Anaheim CA USA March2006 paper NWD6 101109OFC2006215670

[5] YONEYAMA M AKAZAWA Y lsquoBit error rate degradation dueto a dip in the frequency response of an equalizingamplifierrsquo J Lightwave Technol 1999 17 (10)pp 1766ndash1773 doi 10110950793748

[6] JORGESEN D MARKI CF ESENER S lsquoImproved high passfiltering for passive optical networksrsquo IEEE PhotonicsTechnol Lett 2010 22 (15) pp 1144 ndash 1146 doi101109LPT2010205114436

[7] KARLSSON M BRENTEL J ANDREKSON PA lsquoLong-termmeasurement of PMD and polarization drift in installedfibersrsquo J Lightwave Technol 2000 18 (7) pp 941ndash 951doi 101109JLT2004825330

[8] LIN Q AGRAWAL GP lsquoStatistics of polarization dependentgain in fiber-based Raman amplifiersrsquo Opt Lett 2003 28(4) pp 227ndash229 doi 101364OL28000227




ISSN 2041-5923

Compact inductorless CMOS low-noiseamplifier for reconfigurable radioAbdelhalim Slimane1 Fayrouz Haddad2 Sylvain Bourdel3

Sid Ahmed Tedjini-Baıliche1 Mohand Tahar Belaroussi1

Trabelsi Mohamed4 Herve Barthelemy2

1Centre de Developpement des Technologies Avancees Cite du 20 Aout 1956 Baba Hassen Algiers Algeria2IM2NP Aix-Marseille University Technopole Chateau-Gombert Marseille France3IMEP-LAHC Universite de Grenoble Minatec 38000 Grenoble France4Ecole Nationale Polytechnique drsquoAlger Avenue Hacen Badi ndash EL Harrach Algiers AlgeriaE-mail aslimanecdtadz

Abstract A compact reconfigurable CMOS low-noise amplifier (LNA) is presented for applications in DCS1800UMTS WLAN-bg and Bluetooth standards The proposed LNA features first a current reuse shunt-feedbackamplifier for wideband input matching low-noise figure and small area Secondly a cascode amplifier with atunable active LC resonator is added for high gain and continuous tuning of bands Fabricated in a 013 mmCMOS process the measured results show 20 dB power gain 35 dB noise figure in the frequency rangeof 18ndash24 GHz return losses S11 and S22 lower than 212 and 214 dB respectively with a moderate IIP3 of2118 dBm at 24 GHz It consumes 96 mW from a 12 V supply voltage while occupying an active siliconarea of only 0052 mm2

IntroductionSpectacular recent developments in wireless technology arepushing the boundaries of communication standards andplacing higher requirements on next generation wirelesssystems Nowadays RF front ends based on parallelnarrowband paths the single wideband path and the singlemulti-band path are the only ways to implement multi-standard architectures [1ndash3] However designspecifications for such architectures increase the integrationissues of radio sections leading to high costs and highpower consumption Therefore the design ofreconfigurable radios based on maximum sharing of RFbuilding blocks and a minimum of passive inductors isexpected to deal with multi-bandmulti-standardtransceivers regarding cost and energy saving In thiscontext recent works are still focusing on low-noiseamplifier (LNA) reconfigurability to meet multi-standardrequirements In [2ndash4] the designed LNAs employedeither discrete or continuous tuning to cover different

standards whereas passive inductors caused a large siliconarea In contrast inductorless LNAs are also investigatedfor efficient silicon area purposes However theirdrawbacks are either poor performance in multi-bandLNAs [5] or the selectivity issue in wideband LNAs

In this Letter a compact reconfigurable CMOS LNA isproposed for a single-path multi-band front-end usingcontinuous tuning for low sensitivity to process variationsThe following Section describes the LNA circuitMeasurement results and comparisons are then provided

Reconfigurable LNA designFig 1 illustrates the proposed reconfigurable LNA thatsatisfies all the above-specified standards On the basis ofthe two cascaded amplifiers for high gain purposes the firststage offers a good trade-off between wideband matchingand low-noise figure whereas the second is more selectiveby including a tunable active inductor load



Shunt-feedback current reuseamplifierThe shunt-feedback amplifier is a topology that offerswideband input matching with the advantages of designsimplicity relatively low-noise figure and small die areacompared to those based on passive inductors Theamplifier analysis in terms of input impedance (Zin) gain(Av) and noise factor (F ) is given as follows

Zin = RF + RL

1 + gmRL

(1)

Av = RL(1 minus gmRF)

RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2

+ middot middot middot

middot middot middot + 1

Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)

where parameters g and d depend on the process technologyand a is a transistor ratio of transconductance toconductance

From (2) and (3) maximising the transconductance gm

and the load resistance RL is necessary to achieve high gainand a low-noise figure for the first stage according to theFriis formula Therefore a current reuse topology is usedinstead of a conventional shunt-feedback to increase theequivalent gm to gmM1

+ gmM2and the equivalent RL to

RdsM1||RdsM2

for a required DC current As a result thenoise factor becomes completely dependent on gm and RFFurthermore the design constraints are relaxed by solving(1) for the input return loss S11 210 dB which leads tochoosing RF for a good trade-off between high gm and alow-noise figure

Cascode amplifier with activeinductor loadAs shown in Fig 1 to provide high gain and good outputinput isolation the second amplifier uses a cascodetopology formed by two stacked transistors M3 and M4 andsupplied through transistor M5 The cascode amplifier isloaded by the tunable active inductor based on a gyrator-Cnetwork It consists of two connected back-to-backtransconductors gmA implemented by the differential pair(M7 M8) and gmB is implemented by the common sourceamplifier M10 To increase the quality factor a resistanceR1 is inserted between the two transconductors Thus theequivalent inductance and series resistance of the activeinductor are approximated as follows [6]

L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)

where Cgs10 is the gatendashsource capacitance of the transistorM10 roA and CoA are respectively the output resistanceand capacitance of the differential pair transconductor Thetransistor M15 and the resistor R2 are used to biasthe active inductor whereas M12 and M9 mirror a ratio ofthe reference current in M10 and the differential pair(M7 M8) respectively A controlled varactor Cvar is alsoadded at the source of the differential pair (M7 M8) totune the quality factor against frequency

Experimental results and discussionIn addition to the previous simulated results in [7] the LNAprototype is fabricated using 013 mm CMOS technologyThe micrograph of the die is shown in Fig 2 where thetotal die area including pads and decoupling capacitors is0165 mm2 However the active area including decouplingcapacitors is 0052 mm2 (250 times 210 mm) The fabricatedLNA includes a source follower as an output buffer to drivethe 50 V input impedance of the network analyser

Figure 1 Schematic of proposed reconfigurable multi-standard LNA

Figure 2 Micrograph of fabricated multi-standard LNA



Fig 3 depicts the applied Vg effect on gyrator efficiencyFor Vg frac14 02 V a flat power gain is observed indicatingthat the gyrator is still off However by increasing graduallyVg the power gain increases to reach its maximum at Vg frac14

06 V meaning that the gyrator achieves its high qualityfactor at this voltage Then for a fixed Vg frac14 06 V asshown in Fig 4a the power gain is tuned from 18 to24 GHz by a continuous tuning of the varactorrsquos voltage(Vctrl) where more than 22 dB is performed at 18 GHz

for Vctrl frac14 0 V However when Vctrl increases theinductorrsquos quality factor decreases leading to a lower powergain of 206 dB at 24 GHz for Vctrl frac14 12 V Owing tothe impact of the first stage on noise contributions themeasured NF is not affected by the Vctrl variations where itslightly varies between 32 and 35 dB in the frequencyrange of 18ndash24 GHz as shown in Fig 4b

In Fig 4c the current reuse stage in charge of inputmatching achieves a good S11 lower than 212 dB for allbands of interest A good output matching is also ensuredby a low S22 214 dB As a result the continuous tuningby Vctrl increases the robustness of the LNA against processvariations and allows an easy frequency shift to the targetstandards By including the buffer nonlinearities the IIP3is also measured across the band showing moderate valuesfrom 216 to 2118 dBm as given in Fig 4d From asupply voltage of 12 V the LNA power consumption isabout 138 mW including the buffer and 96 mW without it

Operating as narrowband LNAs at one frequency f0 at atime the LNA figure-of-merit (FoM) is given as follows [8]

FoM[GHz] = Power Gain[abs] middot IIP3[mW] middot f0[GHz]

(F minus 1)[abs] middot PDC[mW]

(6)

In Table 1 a performance summary of the recent published

Figure 4 Measured performances for Vg frac14 06 V and different values of Vctrl

a Power gainb Noise figurec S11 and S22

d IIP3

Figure 3 Measured power gain for different values of Vg



multi-standard CMOS LNAs is given for comparison TheFoM of the proposed LNA is as good as the otherreported FoMs Besides the relevant parameter ofperformance and cost-saving design is the ratio of FoM tochip area [9] Occupying an active silicon area of only0052 mm2 the proposed LNA demonstrates a better ratioof FoM to chip area with a good robustness against theprocess variations by means of continuous tuning

ConclusionIn this Letter a reconfigurable CMOS LNA is implementedin an active silicon area of only 0052 mm2 On the basis ofcascaded amplifiers and an active LC load the fabricatedLNA performs more than 20 dB of gain and 35 dB ofnoise figure while consuming 96 mW from a 12 V supplyvoltage in the frequency range of 18ndash24 GHz One canconclude that the proposed LNA could be a goodcandidate to address DCS1800 UMTS WLAN-bg andBluetooth standards for low power and low cost

References

[1] LISCIDINI A BRANDOLINI M SANZOGNI D ET AL lsquoA 013 mmCMOS front-end for DCS1800UMTS80211b ndash g withmultiband positive feedback low-noise amplifierrsquo IEEEJ Solid-State Circuits 2006 41 pp 981ndash989

[2] PARK J KIM S-N ROH Y-S ET AL lsquoA direct-conversionCMOS RF receiver reconfigurable from 2 to 6 GHzrsquo IEEETrans Microw Theory Tech 2010 58 pp 2326ndash2333

[3] FU C-T KO C-L KUO C-N ET AL lsquoA 24 ndash 54 GHz widetuning-range CMOS reconfigurable low-noise amplifierrsquoIEEE Trans Microw Theory Tech 2008 56 pp 2754ndash2763

[4] EL-NOZAHI M SANCHEZ-SINENCIO E ET AL lsquoA CMOS low-noise amplifier with reconfigurable input matchingnetworkrsquo IEEE Trans Microw Theory Tech 2009 57pp 1054ndash1062

[5] ANDERSSON S SVENSSON C lsquoA 750 MHz to 3 GHz tunablenarrowband low-noise amplifierrsquo 23rd NORCHIP Conf2005 pp 8ndash11

[6] ABDALLA MAY PHANG K ELEFTHERIADES GV lsquoPrinted andintegrated CMOS positivenegative refractive-index phaseshifters using tunable active inductorsrsquo IEEE TransMicrow Theory Tech 2007 55 pp 1611ndash1623

[7] SLIMANE A BELAROUSSI MT HADDAD F ET AL lsquoAreconfigurable inductor-less CMOS low noise amplifier formulti-standard applicationsrsquo IEEE 10th Int New Circuitsand Systems Conf (NEWCAS) 2012 pp 57ndash60

[8] BREDERLOW R WEBER W SAUERER J ET AL lsquoA mixed-signaldesign roadmaprsquo IEEE Des Test Comput 2001 18pp 34ndash46

[9] CHEN H-K CHANG D-C JUANG Y-Z ET AL lsquoA compactwideband CMOS low-noise amplifier using shunt resistive-feedback and series inductive-peaking techniquesrsquo IEEEMicrow Wirel Compon Lett 2007 17 pp 616ndash618

Table 1 Performance summary and comparison with state-of-the-art reconfigurable CMOS LNAs

Reference [1] [3] [4] [5] This work

Technology (mm) 013 013 013 018 013

Tuning Discrete Discrete Cont Cont Cont

Passive inductors Yes Yes Yes No No

Frequency (GHz) 182124 24ndash54 19ndash24 075ndash3 18ndash24

S21 (dB) 23ndash29a 22ndash24 10ndash14b 24ndash28 206ndash221

NF (dB) 52ndash58 22ndash31 32ndash37 48ndash10 32ndash35

III3 (dBm) 2750248 216ndash 221 267 224 216ndash118c

Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306

a LNA + mixer gain b Gain voltage c Including buffer nonlinearities d FoM at f0 frac14 24 GHz



IET Digital LibraryBuilding the best in engineering research and information

wwwietdlorg

Keep at the forefront of research by accessing a portfolio of information in a range of engineering fields on the recently re-launched IET Digital Library

Improved navigation and a refreshed appearance makes it even easier to find the essential engineering research that you need in the format that you need it

Includes access to

180000 research journal and magazine articles 3000 eBook chapters 1300 conference publications Over 70000 archive articles dating back to 1872

To build your unique package of engineering and technology research and to take a look at our improved site visit -

The Institution of Engineering and Technology is registered as a Charity in England amp Wales (no 211014) and Scotland (no SC038698) The IET Michael Faraday House Six Hills Way Stevenage SG1 2AY UK

IDL Research Information Advert 210x297indd 1 24012013 094322

wwwtheietorgukjournals

Wersquove been listening

We always keep our ear to the ground and we understand that the communications industry has specifi c information needs So wersquove created a unique package of IET Journals to meet them

Our highly regarded Journals contain a wealth of the latest high quality peer reviewed research Wersquove brought some of the most closely associated titles together in a meaningful package to create a cost-effective addition to your libraryrsquos list

Our expertsrsquo package covers theory and application-based papers on

systems networks and applications involving line mobile satellite and optical technologies for telecommunications state-of-the-art cryptographic and computer security microwave and RF circuits antennas and related systems the theory and practical application of systems and signal processing for radar sonar radiolocation radionavigation and surveillance the growing fi eld of wireless sensor networks and distributed systems the fundamental developments and advancing methodologies to achieve higher performance optimized and dependable future networks

Available online or in print Pay just 20 more to receive them in both formats If you wish you can add our renowned publication Electronics Letters to the package

Yoursquove told us you need the complete package

The Institution of Engineering and Technology is registered as a Charity in England amp Wales (no 211014) and Scotland (no SCO38698) The IET Michael Faraday House Six Hills Way Stevenage SG1 2AY UK

Vol 3 | Issue 1 | ISSN 1751-8725

January 2009

IET Microwaves Antennas amp PropagationINSIDE State-of-the-art papers dedicated to coverage on microwave and RF circuits antennas and related systems

Published by The Institution of Engineering and Technology

Journals_visuals_final_print readyindd 3 30908 153839

IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009

INSIDE Systems networks and applications involving line mobile satellite and optical technologies for telecommunications



IET Information Security

Vol 3 | Issue 1 | ISSN 1751-8709

January 2009

INSIDE State-of-the-art cryptographic and computer security papers from expert authors worldwide


IET Radar Sonar amp Navigation

INSIDE Papers covering the theory and practical application of systems and signal processing for radar sonar radiolocation radionavigation and surveillance

Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today

IET Wireless Sensor Systems

INSIDE Latest research in wireless sensor networks and distributed systems

Vol 1 | ISSN 2043-6386

March 2011

Published by the Institution of Engineering and Technology wwwietdlorgIET-WSS

IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012

Published by the Institution of Engineering and Technology wwwietdlorgIET-NET

INSIDE Covering the very best in communication networks research

IET Listening advertindd 1 06102011 1529

IET eBooksAdding value to your research

Covering 12 series IET books and eBooks have developed a strongreputation in the technical engineering professions providing newperspectives and developments in emerging subject areas

Subject areas include Power Renewable EnergyControl Telecommunications Circuits andElectromagnetic Waves

High quality professional and technical content DOIs to chapter level COUNTER-compliant usage statistics Searchable at book and chapter level Perpetual and subscription access packages No DRM restrictions Available as full or series collections frontback

amp archive lists with MARC21 records

wwwtheietorgeBooks


300+

titles available

Full Page Advertindd 1 18072012 095507

Multimedia Communications Network

About the NetworkThe Multimedia Communications Network is a free to join global community bringing together broadcasting communication and technology professionals interested in electronics and creating dynamic media for different devices in multiple locations

The network enables and supports transformational behaviour through technology evolution and revolution across government and public services business and commerce education and entertainment

If you work or have an interest in broadcasting or telecommunications then come and join our Community

Supporting sector

wwwtheietorgmultimedia

IET MMC A4 Advertindd 1 21082014 1509

Front cover_hi-res IET 2014 IBC FINAL

Inside Front cover_ DTG_IETbestof_A4-3mm4

Advert 1 _053-054_Digital_Library_page

Advert 2_025-026_Communications Package

Inside Back cover_IET eBooks_FP_FINAL

Back cover A4 MMC

Find out more about the DTG and see demonstrations from our winners in

Hall 5A09

wwwdtgorguk DigitalTVGroup

DTG

television

innovation

awards

Partnering UK SMEs for a future of TV technology showcase

For a second year Digital TV Group UKTIKTN and the IET have given four SMEsexhibition space at IBC to promote

UK innovation

Come and see the winning innovations that Come and see the winning innovations that will guide the future of TV including

-Viewer facial recognition-Sound triggered content delivery

-Big data analytics-Instantaneous video delivery to social media

DTG Television Innovation Award Winners 2014

Introduction 1

Editorial 3











DG Brooks 22


TG Edwards 30




doesnrsquot apply

KC Noland 44










D Piehler 70



















Who we areIBC


The IET





ISSN 2041-5923

Editorial




















ISSN 2041-5923




















System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC

Introduction 1

Editorial 3











DG Brooks 22


TG Edwards 30




doesnrsquot apply

KC Noland 44










D Piehler 70



















Who we areIBC


The IET





ISSN 2041-5923

Editorial




















ISSN 2041-5923




















System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC













Who we areIBC


The IET





ISSN 2041-5923

Editorial




















ISSN 2041-5923




















System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC

ISSN 2041-5923

Editorial




















ISSN 2041-5923




















System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC







ISSN 2041-5923




















System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC







System architecture











FFT 32k 8k

Tg 1128 116









Network set-up





























Mobile measurements

































References





ISSN 2041-5923
























ISSN 2041-5923




























































References








ISSN 2041-5923

















ISSN 2041-5923






















dagger Black level















Colour Volume












1 F-stop 2 F-stops












Colour Quantization


























































REFERENCES


















ISSN 2041-5923

































720p 1080i (F1) 1080i (F2)





















Hardware











Server software












Results






References















ISSN 2041-5923






























Fascinate system



iCaCoT System






















References










ISSN 2041-5923


















t = rn (1)










































References






















ISSN 2041-5923



























Rec BT2020 RecBT709


Red primary (R) 0708 0292 0640 0330

Green primary (G) 0170 0797 0300 0600

Blue primary (B) 0131 0046 0150 0060






















⎡⎣

⎤⎦













Interface design











250 N





Signal mapping




Standardisation











Contrast 5000 1

Bit depth 12 bits



CIE x CIE y



















References





















ISSN 2041-5923



















Test material




2 RAVC(i)


Laboratory Setup

















Analysis






MOS BD-rate











Book 266


Manege 256

Traffic 258


Calendar 252

SVT15 269

sedofCropped 253

UnderBoat1 268


BT709Parakeets 266


SVT01a 273

SVT04a 236


Anemone 242


Ducks 272


Average 2587













References
























ISSN 2041-5923


































References












ISSN 2041-5923































3[ ]







3 996 00130 09979 002 008

6 987 00247 09930 011 051

9 976 00397 09861 025 149

12 963 00546 09777 048 378






References












ISSN 2041-5923













Zin = RF + RL

1 + gmRL

(1)


RF + RL

(2)

F ≃ 1 + 1

RsRF

1

gm

+ Rs

( )2



Rs

1 + Rs

RF

( )2 1

g2mRL

+ g

agm

+ da

5gm

( )(3)





RdsM1||RdsM2



L ≃ 2

gmAgmB

Cgs101 + R1

roA

( )+ CoA

( )(4)

Rs ≃2

gmAgmB

1

roA

minus v2CoACgs10R1

( )(5)














(6)




d IIP3






References












Technology (mm) 013 013 013 018 013







Vdd (V) 12 1 12 18 12

PDC (mW) 24 46 17 42 96

Area (mm2) 06 049 0083 009 0052

FoM (GHz)d 258 141 022 0005 159

FoMarea 43 288 27 005 306





wwwietdlorg



Includes access to














Vol 3 | Issue 1 | ISSN 1751-8725

January 2009




IET Communications

Vol 3 | Issue 1 | ISSN 1751-8628

January 2009





Vol 3 | Issue 1 | ISSN 1751-8709

January 2009





Vol 3 | Issue 1 | ISSN 1751-8784

January 2009


Subscribe today



Vol 1 | ISSN 2043-6386

March 2011


IET Networks

Vol 1 | Issue 1 | ISSN 2047-4954

March 2012









wwwtheietorgeBooks


300+

titles available






Supporting sector








Back cover A4 MMC

the best of iet and ibc 2014 · the best of iet and ibc, 2014, vol. 6, pp. 1–2 2 & the...

Documents