`
`A NEW GENERATION OF MPEG-2 VIDEO ENCODER ASIC & ITS APPLICATION TO NEW
`TECHNOLOGY MARKETS
`'
`
`M Irvin, T Kitazawa, T Suzuki
`
`Sony Broadcast & Professional Europe , UK
`
`ABSTRACT
`
`The multimedia era is upon us. The traditional methods used for the storage, signal processing and distribution of video,
`audio and data are going through significant change. This change is being fuelled by recent advances in the area of bit-
`rate reduction technology and in particular the publication of an effective world-wide standard for the compression of
`video and audio components, the MPEG—2 specification (1).
`
`This paper considers the complexities involved in performing video compression encoding. It introduces newly
`developed Sony programmable ASIC’s for performing real time compression encoding of a video source image in
`compliance with the MPEG-2 Main Profile @ Main Level standard. It details the architecture of this newly developed
`chipset including features that have been implemented within me chipset to identify image complexities, resulting in
`efficient video compression encoding with high picture quality.
`
`The paper describes how this video compression chipset has been integrated into a new generation MPEG-2 Encoder
`product from Sony and identifies a number of new technology application areas which are being addressed by this new
`product.
`
`INTRODUCTION
`
`Video compression technology has made significant
`advances in recent years and a succession of
`compression formats have emerged. Figure 1 identifies
`a number of such compression formats and compares
`their achievable data rate range.
`
`Data Rate
`(W)
`50 M
`
`15M
`
`1.5M
`
`384 K
`
`64 K
`
`
`
`E
`=‘
`
`:9
`E
`E
`
`a
`E
`E
`
`8
`&
`
`E
`g
`E
`(5
`E
`
`E
`3
`§
`‘7‘
`E‘
`4
`
`Figure 1 Compression Formats
`
`International Broadcasting Convention, 12—16 September 1996
`Conference Publication No. 428, © IEE, 1996
`
`Although each codec has its application area, one of the
`most prominent is the MPEG-2 Main Profile @Main
`Level because of its widely covered data bit-rate and
`sophisticated reduction method.
`
`MPEG-2 VIDEO COMPRESSION TECHNOLOGY
`
`The compression technique adopted by the Motion
`Picture Experts Group for the MPEG—2 family of
`profiles and levels is designed around an asymmetrical
`process. The nature of this asymmetrical process is
`designed to concentrate the required complex image
`analysis and processing to the compression encoding
`stage, whilst realising a significantly less complex
`decompression decoding stage.
`
`By adopting this asymmetrical process, a cost effective
`decoding device can be achieved, although this does
`result in a cost penalty for the encoding device in
`comparison. However when considering atypical
`application area for MPEG—2, it can be seen that the
`number of decoding devices required far exceeds the
`number of encoding device by many thousand or
`hundred of thousands. The encoding device cost penalty
`can therefore be sustained.
`
`The MPEG-2 specification is defined with respect to the
`decoding process and uses what is termed a ‘System
`Target Decoder’. It is therefore open to manufacturers
`of MPEG-2 encoding systems to utilise to their best
`advantage techniques to improve the encoding process,
`as long as the resultant bit-stream is both syntactically
`correct in terms of MPEG-2 and that the resultant bit-
`
`stream can be decoded by the MPEG-2 defined System
`Target DCCOdCF.
`
`1
`
`LG 1020
`
`LG 1020
`
`1
`
`
`
`392
`
`We can summarise this by saying that in terms of picture
`quality, there is little perceived difference between
`different manufacturer supplied decoder devices.
`Moreover it is the encoding techniques adopted by a
`particular manufacturer which achieve the overall
`resultant picture quality and hence positions a
`manufacturers MPEG-2 encoding device against its
`competitors.
`
`MPEG-2 VIDEO ENCODING CONSIDERATIONS
`
`Within a video compression application, the source
`picture to be compressed can be found to have a number
`of characteristics. These characteristics are dependent
`upon:-
`-
`
`The method used to capture or record the
`original source image, and
`
`o
`
`The content of the original scene or image that
`was captured
`
`By identifying these characteristics, a video compression
`encoding device can exploit the rich set of tools that the
`MPEG-2 syntax provides to produce are more efficient
`coded image.
`
`Image Capture -The method used to capture the
`original source image will determine the nature of the
`picture type, for example, an image captured using a
`film type camera will determine a progressively scanned
`frame, whereas an image captured using a video type
`camera may determine an interlaced frame consisting of
`two fields.
`
`In addition to the method used to capture the original
`image, we must also consider how the original image
`was recorded and ‘transported’ throughout the
`production and post production chain.
`In the case of an
`image that was captured using a film type camera, it is
`possible that the film image sequence may have been
`transferred to a video based format which has both a
`different frame rate and frame structure.
`
`
`
`FilmFrame# i—1
`2
`I
`3
`4J 5
`
`
`
`Video Field Sequence
`
`1 2 1I1 2 1 2 1 2I2
`
`
`
`
`VideoFrame#'1‘2l3l4 5'6
`
`
`Figure 2 NTSC 3:2 Pulldown Sequence
`
`In NTSC television based countries, it is common to
`transfer 24 frame per second film based material to a
`video based storage device, which characterises both an
`interlaced frame picture structure and a field rate of
`59.94 Hz. The resultant transfer process is referred to as
`
`‘3:2 pulldown’. This technique is detailed in Figure 2,
`in which it can be seen that the progressive frame
`structure is firstly converted to an interlaced field
`structure and secondly each fifth video field is a repeat
`field to achieve the desired overall video field rate.
`
`In PAL based countries, the situation is slightly easier,
`as the conversion process involves a transfer from 24
`frame film to an interlaced 50 field based video system.
`This is commonly achieved by increasing the replay rate
`of the film source material to 25 frames per second and
`then performing a progressive to interlaced scan
`conversion, resulting in two interlaced video fields
`which represent the original single film frame.
`
`One cause for concern however, in PAL based film
`transfers, is the video field alignment accuracy. It is
`possible within the film transfer process, that the
`resultant interlaced video field to frame relationship is
`disturbed. This causes the odd interlaced video field
`and the even interlaced video field in a video frame to
`
`have originated from different consecutive film frames.
`This error process is commonly referred to as ‘PAL field
`fllp’.
`
`Having identified the source picture image type, the
`MPEG-2 syntax provides tools which can be utilised in
`a manner that efficiently encodes the input video
`sequence. For example at the macroblock layer, the
`flags representing dct_type and prediction can be
`described as either field or frame based. In the case that
`
`an input video sequence has been 3:2 pulldown
`processed, a video encoding device which correctly
`identifies the 3:2 sequence can use the repeat_first_f1eld
`flag at the picture coding layer and hence avoid the need
`to encode redundant fields, leading to increased
`encoding efficiency.
`
`Image Content -The image content of the video
`sequence to be encoded can have a drastic effect on the
`overall picture quality. In everyday use, we can see
`video material which contains images representing
`movie film, sports events, commercials and news for
`example. Each image type has varying characteristics of
`image complexity placing differing demands on the
`video encoding process. Complexities can be dependent
`on scene content, such as fast moving sports images or
`strobe lighting, or on the image sequence structure
`itself, such as scene changes or fade-up and fade-down
`effects.
`‘
`
`It is important that a video encoding device is able to
`dynamically react to these differing image complexities
`and keep encoding artifacts to a minimum. By
`proficient use of the MPEG-2 coding syntax and the
`resultant effective use of macroblock parameters,
`derived from inspecting each macroblock statistical
`characteristic, this can be achieved.
`
`In the case of Constant Bit Rate (CBR) applications.
`additional constraints apply to the video encoding
`
`2
`
`
`
`393
`
`process. The document TM-S (2), Test Model 5 of the
`MPEG-2 standard gives guidance in this area. TM—5
`determines three strategies to be adopted to balance the
`resultant video encoded bit—rate:-
`
`ENCONT -The ENCONT device performs frame re-
`ordering, motion estimation control, picture statistics
`and adaptive control. ENCONT can also perform the
`following operations:-
`
`0
`
`0
`
`0
`
`By determining each picture type I, P, or B
`complexity and the balance of bit allocation by
`feedback from the last picture of the same type.
`
`The use of a quantisation index, which is
`controlled each macroblock using feedback
`from the Variable Length Coding (VLC)
`buffer.
`
`By controlling the macroblock quantisation to
`improve picture quality dependent on
`macroblock activity.
`
`It is worth noting however that TM—S does not guarantee
`Video Buffer Verifier (VBV) compliance. Therefore to
`prevent underflow or overflow on VBV, zero stuffing
`and skip macroblock or skip picture syntax code is
`usually used.
`
`SONY MPEG-2 VIDEO ENCODING SOLUTIONS
`
`For MPEG-2 based applications, Sony has developed a
`two chip Application Specific Integrated Circuit (ASIC)
`for real time video encoding. The devices use 0.35
`micron fabrication technology with a gate count
`approaching ] million gates per device. This has
`achieved low power consumption and small size.
`
`Table 1 details further the Sony device specifications.
`
`0 Automatic detection of a 3:2 pulldown
`sequence and effective allocation of the
`repeat_first__field flag in the picture coding
`header.
`
`0
`
`In the case of 625/50 video line standard
`
`operation, ENCONT can detect and
`compensate for PAL field flipped telecine film
`sources materiall.
`
`0 Automatic scene change detection, to improve
`picture quality by dynamically changing picture
`type and bit allocation.
`
`ENCORE -The ENCORE device is responsible for
`Discrete Cosine Transform (DCT), Quantisation,
`Variable Length Coding (VLC) and local decoding.
`The on chip RISC processor within ENCORE is
`responsible for both constant and variable bit-rate
`control, including VBV compliance and has the
`capability to improve upon TM—S rate control. The
`processor is programmable to insert all of the required
`MPEG-2 header syntax.
`
`The chipset has selectable external memory size, such
`that for low_delay applications it can minimise memory
`size operation to 6 MB total.
`
`The chipset can encode MPEG-2 Main Profile @ Main
`Level up to 15 Mbps, not only in real time with constant
`bit-rate, but also as a two pass encoding process with
`variable bit—rate as required for Digital Video Disk
`(DVD) authoring.
`
`In other MPEG-2 application areas, such as Digital
`Video Broadcasting (DVB), both constant bit—rate and
`dynamically variable bit-rates, for Statistical
`Multiplexing operation can be supported.
`
`By implementing a RISC processor inside each encoder
`chip, a variety of MPEG-2 applications can be realised
`by updating the executable microcode on the RISC core.
`
`The Sony BDX—EIOOO MPEG-2 Encoder is the first
`product to feature this newly developed Sony MPEG-2
`encoding chipset ENCONT and ENCORE. The re-
`programmable nature of the encoding chipset at the
`heart of the BDX-ElOOO, gives provision to address a
`number of application areas with a single product
`development. This paper will now highlight some of the
`new application areas which are being addressed by the
`BDX—EIOOO.
`
`Technolo;
`
`Gate count
`
`Packa 1 e
`
`Power consum I tion
`
`ENCONT
`
`ENCORE
`
`0.35
`
`910 K
`
`0.35
`
`910 K
`
`304 SQFP
`
`208 SQFP
`
`2.5 V
`
`0.8W
`
`2.5 V
`
`0.8W
`
`'
`
`External DRAM
`
`4 ~ 32 MB
`
`2~8MB
`
`Table 1 Specification of Chipset
`
`The chipset consists of two devices designated the
`names ENCONT and ENCORE whose architecture is
`
`shown in Figure 3 . The ENCONT, encoder control
`device performs input signal adaptation and motion
`estimation, whereas ENCORE, the encoding core
`performs the compression process.
`
`In designing the chipset, a hybrid architecture has been
`adopted. High speed and stereotype signal processing is
`assigned to hardware and complex algorithmical
`processes such as adaptive control and rate control are
`assigned to programmable RISC processors.
`
`3
`
`
`
`APPLICATIONS OF MPEG-2
`
`Program Archive
`
`394
`
`Digital Video Broadcasting
`
`Digital Video Broadcasting (DVB) is the European
`standard for next generation programme transmission.
`The standard was developed with co-operation from
`over 150 members of a consortium representing a broad
`spectrum of the broadcast industry.
`
`The DVB standard encompasses all elements of the
`broadcast chain, from studio playout to home reception.
`Source programme information, such as video, audio
`and teletext information is transmitted in a digital form.
`
`To achieve efficient use of spectrum bandwidth, the
`DVB committee have endorsed the use of the MPEG—2
`
`compression standard and in particular Main Profile @
`Main Level video compression. The utilisation of
`compression enables a number of programme channels
`to be transmitted in the same spectrum space previously
`reserved for a single analogue channel.
`
`,
`
`The key elements of a Digital Video Broadcasting
`system are highlighted in Figure 4. Within the system,
`source programming is delivered from the programme
`playout area to the input of a number of MPEG-2
`encoders. A single MPEG-2 encoder is designated to
`process the component parts of a single programme
`channel.
`
`The resultant compressed bit-stream output from each of
`the MPEG-2 encoders is combined in to a single data
`stream by a Multiplexing device, the output of which is
`subsequently transmitted across a network. The DVB
`have currently defined four network types for the
`distribution of compressed programme, satellite, cable,
`terrestrial and microwave.
`
`To complete the DVB transmission chain, a decoder unit
`is required at the consumers home to receive the
`transmitted signal. This device, referred to as an
`Integrated Receiver Decoder (IRD) performs signal
`adaptation, demodulation and MPEG-2 compliant Main
`Profile @ Main Level decoding.
`
`The Sony BDX-EIOOO MPEG—2 Encoding unit, which
`features the new generation video ASIC’s described in
`this paper is fully DVB compliant. The unit has a
`flexible architecture to meet differing broadcasters
`requirements, making it an ideal product for DVB
`application.
`
`The unit is compact in size and is designed to process all
`of the component parts of a single programme channel,
`including video, up to 8 audio channels and private data
`such as DVB subtitles. The device can be controlled
`
`and configured from an external PC or Workstation
`using time schedule information through either Ethernet
`or RS-232C connection.
`
`The popularity of the television image within the last 30
`years, has created an explosion in the television
`programme production market. This has resulted in the
`creation of vast libraries containing historical archives
`of programme material. As the popularity of television
`continues to grow, particularly as we are now moving
`into the age of digital television, there is a constant need
`for programme material to fill airtime schedules.
`
`One of the potential sources for this programme material
`is from the program archive collections which contain
`many classic television programme produced within the
`most popular period of television history to date.
`
`Unfortunately, many of these programme’s were
`produced on storage media, which is now becoming
`unstable as it nears the end of its life. To prevent
`irretrievable loss of this precious programme material,
`the archivist must regenerate this material on to a new
`storage media. However this can be a very time
`consuming and costly process as new storage media has
`to be purchased.
`
`The application of MPEG-2 compression to Program
`Archive can bring significant benefits, by offering a
`flexible compression scheme which can be optimised to
`meet picture quality against storage bandwidth
`requirements.
`
`Figure 5 details a Program Archive system, in which a
`number of MPEG-2 Main Profile @ Main Level
`
`Encoders are located at the heart of the system.
`Programme source material is played into the MPEG-2
`Encoders, which is subsequently compressed and stored
`on a new storage device, in this example a data tape
`recorder located in a robotic cart machine.
`
`The archive process is controlled by a higher level
`software application, from which the archivist creates a
`database record of the archive material and determines
`
`the parameters to be used for the compression process.
`
`The Sony BDX—ElOOO MPEG-2 Encoder is ideally
`suited to the Program Archive application, as it supports
`numerous input video and output data interface formats,
`meeting customer requirements for choice of storage
`devices.
`
`Digital Video Disk
`
`The Digital Video Disk (DVD) is the next generation of
`consumer optical media, offering vastly increased
`storage capacity over its predecessors, with the ability to
`store an entire programme movie in digital video format
`on a single disk side.
`
`To achieve this storage capacity, DVD has adopted
`MPEG—2 Main Profile @ Main Level compression. The
`DVD format gives provision for data rates averaging 3.5
`Mbps, with peak demands as high as 10 Mbps. Higher
`bit-rates are assigned to complex pictures and lower bit-
`
`4
`
`
`
`395
`
`rates to simpler pictures using an adaptive variable bit-
`rate process. This process is performed in two stages,
`the first being an evaluation stage to determine picture
`complexity and the second the actual compression
`process.
`
`The newly developed ASIC’s described in this paper
`can support this two stage coding process. Figure 6
`details a Digital Video Disk Mastering system, in which
`the Sony BDX-ElOOO MPEG-2 Encoder product forms
`an integral part.
`
`The source programme material components are passed
`through their respective encoding devices and then
`combined at a multiplexing and disk formatting stage.
`The compressed and formatted disk master data is then
`streamed on to a removable media located inside a data
`
`storage device. This removable media is then forwarded
`to the disk mastering plant, where the optical disks are
`subsequently pressed.
`
`CONCLUSION
`
`This paper has described SONY’s new generation
`MPEG-2 video ASIC’s and their application to new
`technology markets.
`
`High picture quality is achievable in both constant bit—
`rate and variable bit-rate compression modes through
`the use of image type recognition functions and adaptive
`bit—rate control by the video ASIC’s.
`
`The paper described three application areas addressed
`by a new Sony product featuring the new generation
`video ASIC’s, Digital Video Broadcasting, Program
`Archive and Digital Video Disk.
`
`REFERENCES
`
`1. ITU—T Recommendation H.262 I ISO/IEC
`
`13818: 1995, Information technology -Generic coding of
`moving picture and associated audio information.
`
`2. International Organisation for Standardisation,
`ISO/IEC JTCl/SC29/WG11 Coded Representation of
`Picture and Audio information. ISO/IEC
`JTCl/SC29/WG1 1/N0400 Test Model 5 .
`
`ACKNOWLEDGEMENTS
`
`The authors would like to thank their colleagues for
`their contributions and support during the preparation of
`this paper.
`
`5
`
`
`
`
`
`
`
`
`
`5:59:25m>Ee<5:923;SEE<
`
`
`
`
`
`
`
`
`
`pmmmc‘wS—mine/K
`
`
`
`
`
`
`5.22003.52
`
`396
`
`uofimmbmwas.
`
`
`
`43@m2NAME:
`
`meOUZm
`
`523‘.viaEn:
`
`Seesaw2:35
`
`5535::
`
`EdaEgonmm
`
`
`
`
`
`
`
`
`
`
`Eouwhmmew—0632Mama32>Rama9whim
`DEEEG anmhm
`
`wcflmmowuem82>3:39v«Ema
`
`coxmflEmcmta$33.8:
`
`
`
`363%:a388m
`
`
`
`mmmuo<\mcgsutoo
`
`
`
`
`
`
`
`
`
`.mEmEmmmcwE
`
`
`
`Kmntumozm:8wawEEEmSQ
`
`
`
`
`52mm32530
`
`9.200an
`
`
`
`.ogcos.w>EQ<
`
`
`
`.=2@n=2Numis—
`
`.2805
`
`
`
`5E0:850m
`
`wiwwouoi
`
`250203
`
`BEE»:
`
`MM<>>QM<=
`
`Hanson«ax
`
`>2“.335an
`
`>3Eunommm
`
`3::90
`
`8585358
`
`
`
`«5:33;:
`
`3:30032
`
`$28an:
`
`
`
`”9.562.3m
`
`
`
`uo§m€<85%SE
`
`£23m5Mamas.
`
`.2395‘
`
`EmaEbxm
`
`5.883.
`
`
`
`mm..2595
`
`456m5Mama
`
`53335
`
`
`
`35:qu“Eu—v.03?
`
`£5259288SanuE_._=>9:.qu
`
`
`
`
`
`
`
`589$o>EP~<Ewawoumm9.59%
`
`258:5:E5mmoozmaHzoozm28m2&3
`
`nay—002nm 53%?
`
`
`
`6
`
`
`
`
`