`Experiences with Real-Time Streaming of Audio Streams
`
`Karl Jonas
`
`Peter Kanzow
`
`Mathias Kretschmer
`
`German National Research Center for Information Technology (GMD)
`SchloR Birlinghoven, D-53754 St. Augustin, Germany
`EMail: { firstname.lastname} C9gmd.de
`
`Abstract - An increasing number of applications include
`real-time audio streaming over a communication net-
`work. Well-known examples are teleconferencing,
`tele-presentation, Internet Radio live streams and
`on-demand services.
`In cooperation with various broadcasters GMD pro-
`vides live and on-dem,and audio services over the Inter-
`net. Applications are lbased on commercial products as
`well as on own software developments.
`This paper describes some approaches to provide a real-
`time audio service over the Internet, discusses problems
`and experiences and presents an outlook for a streaming
`service in the future.
`
`I. Motivation
`
`An increasing number of applications use local area net-
`works and the Internet for data transmission. In addition to
`the traditional text and graphics transmission, provided
`namely by the World Wide Web (WWW), more and more
`audio and video services emerge. This paper focuses on
`audio services especially for the Internet with its currently
`low bandwidth.
`In cooperation with several German broadcasters, GMD
`has implemented an integrated Internet Radio system which
`uses a WWW-interface to provide worldwide
`a live audio stream (the current radio programme)
`access to archives of audio streams, built of previously
`broadcast content. ]<ere one can find the broadcasts of
`yesterday or all news on a given topic.
`Since radio is very fast and about current affairs, the
`audio archives and additional informations have to be
`updated almost continuously, every minute, every day,
`seven days a week. This happens autornatically to make the
`service efficient to the provider.
`GMD uses audio streaming to implement Internet Radio
`services. Several strearni ng systems for audio (and video)
`exist on the market. This paper examines the features which
`make an audio streaming system suitable for an Internet
`Radio service and discusses how existing streaming systems
`meet these requirements.
`Chapter two compars-s the two major methods to deliver
`audio content over a network - download and streaming. It
`also presents some information about encoding standards
`and transmission protoc:ols.
`Chapter three presents the components of typical audio
`streaming systems and discusses requirements a streaming
`system must fulfill to be used with Internet Radio services.
`
`Chapter four presents two commercial streaming sys-
`tems, Xing Streamworks and RealAudio and discusses how
`they meet or fail to meet the above requirements.
`Chapter five describes running Internet Radio services by
`GMD which can be accessed over the Internet now.
`Chapter six discusses problems and possible improve-
`ments leading to a prototype implementation for a future
`system described in the chapter.
`
`11. Delivering Audio Content Over a Low
`Bandwidth Network
`A. Streaming and Downloading
`In general, audio content can be delivered over a network
`in two ways:
`1. The audio file can be downloaded and then played from
`the local hard disk of the client.
`2. The audio content can be streamed from the server to the
`client who decodes the received packets in real time,
`displays the content immediately and then discards the
`received data.
`The advantages of the downloading method are:
`downloading works with any datarate and allows any
`audio quality one wants to offer.
`the file is transmitted error-free thus no quality reduction
`occurs during transmission.
`But there are major drawbacks of this method of audio
`delivery:
`Download times are extremely long if high audio quality
`is provided over a low-bandwidth network. For example,
`a five-minute-long music clip encoded with 128kbps
`takes half an hour or more to download over a typical
`private Internet access with an effective long-term
`bandwidth of 20kbps. To reduce download times, audio
`quality must be reduced.
`The user has to load the complete file before he can
`listen to any part of it. He cannot preview the file to
`decide whether he is interested in the content.
`There is no possibility to provide a “live” service as
`re-broadcasting a radio programme into the network
`simultaneously with e.g. the terrestrial broadcast.
`The advantages of streaming are:
`The user can listen to the content immediately after he
`has demanded for it. He can fast-forward and listen to
`other parts of the content without waiting for the whole
`file to download.
`It is possible to provide ‘‘live’’ services by encoding the
`audio signal in real time and sending the resulting audio
`data stream immediately to the client.
`
`IEEE Catalog Number: 97TH8280
`
`- SS71 -
`
`ISIE’97 - Guimarks, Portugal
`
`Juniper Ex. 1043-p. 1
`Juniper v Implicit
`
`
`
`The drawbacks of streaming are:
`To stream real time audio data, the transmission line has
`to provide the full bandwidth of the stream during the
`whole transmission period. This imposes a limit on the
`bandwidth and hence the quality of the audio stream.
`Concerning the Internet, the possible bandwidth is very
`low if you want to achieve the vast amount of private
`usei-s connected with a modem and via a provider such
`as Compuserve, America Online (AOL), etc.
`The transmitted stream is very sensitive to network load
`which inay cause lost or delayed data packets. This
`leads to drop-outs in the client audio output.
`To process the required flow control, additional server
`software (a “streaming server”) is needed.
`It has to be considered that the main advantage of down-
`loading, the possible high audio quality even over low
`bandwidth networks, is compensated in practical use by the
`required high download times.
`The drawbacks of streaming, however, can be reduced by
`dynamically adapting the amount of transmitted data to the
`actual capacity of the network connection. To do this, the
`audio data must either be stored in a format that allows
`dropping parts of the data resulting in a “graceful degrada-
`tion” or they must be stored in different formats that the
`server can choose the format best suited for the bandwidth
`of a given connection.
`In conclusion, streaming is the suitable method to offer
`the user a fast access to the content. It is the only solution
`possible to offer a live stream.
`B. HTTP Streaming
`[9]
`(HTTP)
`The
`,,Hypertext Transport Protocol“
`describes how data is transmitted in the World Wide Web.
`HTTP “streaming” tries to reduce some of the drawbacks
`of downloading. The file is downloaded via HTTP (i.e no
`special Streaming Server is involved) but displayed by the
`client immediately when the data arrive. Thus, the user can
`listen to the content immediately, as with streaming. But
`there is no fast forward and no possibility for the system to
`react to insufficient network bandwidth or temporary net-
`work congestions. Data are transported via the reliable
`TCPEP protocol stack, which is slower and less appropriate
`for audio streaming than e.g. UDP (see section D of this
`chapter) .
`C. Audio Formats
`All streaming systems use more or less different percep-
`tual audio codecs to convey as much audio quality as possi-
`ble to the listener. The currently most frequently used
`codecs are Dolby AC-3 based ones (i.e. RealAudio 3.0) and
`MPEG Layer 2Layer 3 implementations (i.e. XING Stre-
`ainworks, Fraunhofer’s L3Enc). GMD additionally uses
`AT&T’s PAC codec. All codec implementations are real-
`time capable on at least one hardware platform (i.e they are
`optimized for Intel Pentium Processors). Hence they can be
`used with Realtime-Audio-Streaming applications
`like
`“MBone Live Radio”.
`The perceived audio quality differs from codec to codec
`and depends on the given bitrate. Real-Audio seems to pro-
`duce the best results at very low bitrates between 8 and 16
`kBit/s. At bitrates up to 112 kBit/s, AT&T’s PAC and
`
`Fraunhofer’s (FhG) MPEG-Layer3 implementation repro-
`duce the original with the lowest noticeable distortion. In
`fact, the difference between the original and the codec out-
`put is smaller the higher the requested bitrate is.
`We will soon examine the differences in the error toler-
`ance of the various codecs focusing on packet losses.
`Additionally, we will test the whole streaming system
`under (simulated) Internet conditions. We will focus our
`work on the system‘s behavior on packet delays und packet
`losses.
`
`D. Transmission Protocols for Audio Streaming
`
`Endsystem applications do not implement all coinmuni-
`cation features, but use exinsting communication protocols
`instead. Typically, a network protocol is used to forward
`datagrams across a network, and a transport protocol is
`used for end-to-end services. The combination of protocols
`is called the protocol stack.
`The typical protocol stack which is used for data trans-
`mission across the internet is TCP [6] or UDP [7] on top of
`IP [jl (fig. 1)
`
`I APPLICATION I B
`
`APPLICATION TI
`
`ROUTER
`
`ETHERNET
`
`ETHERNET
`
`stacks:
`protocol
`for
`Figure 1: Example
`application data is transported via RTP,
`UDP, IP and EtherneUATM
`
`The well-known TCP ’ protocol provides reliable data
`transmission. It acknowledges received data to the sender
`then being able to retransmit lost data. Due to retranxmis-
`sion delays, however, TCP transmission is never continuous
`on an unreliable network. Compared to TCP, unreliability is
`a majorfeature of UDP, lost data is never retransmitted. For
`most streaming applications, packet loss is more acceptable
`than discontinuous presentation.
`The Realtime Transport Protocol [8] can be used on top
`of UDP. It provides additional presentation information
`(timestamps, payload type) for the price of some additional
`transmission and processing overhead.
`A combination of IP, UDP and RTP adds a minimum
`protocol overhead of (20+8+12=) 40 bytes to each data-
`gram. While this is no problem for high quality (= high
`bandwidth) audio, it becomes important for 5 kBit/s
`streams where a 20 ms datagram results in 12 bytes payload
`(typically, payload size is increased, e.g. to 100 ms for very
`low bandwidth. IP/UDP/RTP header compression to a total
`of two bytes per datagram is under discussion)
`
`IEEE Catalog Number: 97TH8280
`
`- SS72 -
`
`ISIE’W - Guimariies, Portugal
`
`Juniper Ex. 1043-p. 2
`Juniper v Implicit
`
`
`
`-
`
`111. Requiremenls to a Streaming System for
`Internet Radio
`
`A. The Components and How They Work Together
`An Internet Radio system as discussed here contains both
`a live stream and automatically updated audio archives. A
`suitable audio streaming system for this service must
`include
`I . permanent realtime digitizing and encoding of the
`(analog) live audio signal into a streamable format;
`2. permanent updating of the audio archives by recording
`parts or the whole encoded live audio data on hard disk;
`3. a streaming server, capable of serving the live stream
`and the recorded streams;
`4. streaming clients which receive and play the audio data
`stream on the client host’s audio system.
`
`STKEAMING CLlEN
`
`IsTREAMING CI-IEN
`
`B. Further Requirernerits
`While working with several streaming systems we
`learned that for practical use the fulfilling of the above
`requirements is not enough. There are additional require-
`ments which are essential for the Internet Radio to work,
`e.g.:
`The audio codec must produce sufficient quality at a
`very low bandwidth. Most Internet users receive a
`permanent data rate of 20kbps or less even if they are
`connected to their Internet provider via ISDN. To reach
`them, an audio stream must not exceed that datarate
`while providing understandable audio quality.
`The streaming server must be capable of serving parts
`of audio files instead of the whole recording only. A
`long recording can thus be divided into parts (which we
`call “clips”) just by creating pointers instead of cutting
`it physically. This makes “cutting”, automation and
`correction much faster and easier.
`
`IV. Commercial Streaming Systems
`
`Today, many systems exist for streaming audio and/or
`video through low bandwidth networks as the Internet. Two
`of the most important ones are Xing Strsarnworks and
`RealAudio by Progressive Networks. We will discuss them
`in this chapter.
`A. Xing Streamworks
`Streamworks by the American company Xing [IO] is
`currently in the version 2.0. It supports both audio and
`video. In this paper, we will focus on audio only.
`Streamworks uses MPEG Layer I and I1 audio codecs
`and a special “Low Bit Rate” (LBR) codec for very low
`bandwidths (< 16kbps)
`The Components
`In terms of the setup described in Fig. 2, Xing offers the
`digitizer and the encoder as a single hardware component
`called the “Xing Audio Transmitter”. Software includes a
`recorder which is called “Capture Tool”, a streaming server
`(“Streamworks server”) and streaming clients (“Stream-
`works Player”). There is no reflector available from Xing.
`The Xirig Audio Transmitter turns out to be a 486-PC
`with a special audio capturedencoder ISA board, an ISA
`network interface board and DOS software.
`The Transmitter digitizes and encodes an analog audio
`signal. The encoded audio data is transmitted as a stream of
`UDP packets into the LAN the Transmitter is connected to.
`Since the Transmitter comes with no hard disk, it boots
`from a floppy disk containing DOS and ATRANS, Xing’s
`encoding control software. A DOS-based, menu-driven
`configuration program lets the user specify parameters like
`the desired audio codec and datarate as well as the Trans-
`mitter’s IP address and the destination IP address and port
`of the output stream.
`The Transmitter’s output stream is meant to be received
`by either the Capture Tool (to be stored and re-broadcast
`later) or by the Streamworks server (to be re-broadcast
`immediately as a live stream).
`
`DIGITIZER
`
`STREAMING SLRVER
`
`&
`
`I
`
`r w L CTOR
`
`RECORDER
`
`
`
`HARD DISK(S)
`
`Figure 2: Components of a typical Streaming
`System
`
`Fig. 2 shows a typical setup of a streaming system for an
`Internet Radio service. The audio source (e.g a radio tuner)
`delivers the content to be re-broadcast in the net. Its output,
`an analog audio signal, is digitized and encoded in real time
`by the digitizer (or analog-digital converter, ADC) and the
`encoder (which can be hardware or software).
`The output of the encoder is compressed digital data in a
`streamable format. The reflector just doubles this stream
`and passes the first to the streanzing server. The streaming
`server transmits the data to the listening streaming client(s),
`if there are any currently demanding the live stream. Other-
`wise, the data are discarded.
`The second data stream is passed to the recorder. The
`recorder stores the data on hard disk according to a sched-
`ule which determines file names, start and end times. The
`recorder thereby fills {he archives with audio files of differ-
`ent length and names, depending on which content is meant
`to be stored. We call these audio files “recordings”.
`Whenever a streaming client demands a stored audio
`stream, the streaming server reads the according audio data
`from hard disk and transmits them to the client.
`
`IEEE Catalog Number: WTH8280
`
`- ss73 -
`
`ISIE’97 - Guimaries, Portugal
`
`Juniper Ex. 1043-p. 3
`Juniper v Implicit
`
`
`
`The Capture Tool receives and stores the Transmitter’s
`output stream as a recording. The tool takes two parameters
`- duration of the recording and the recording’s file name.
`The Streamworks server, which is available for several
`UNIX systems and Windows NT, receives the Transmitter’s
`output stream and serves it as a live stream via the Inter-
`nethtranet to listening Streamworks clients. The server is
`also capable of serving existing recordings as audio
`streams. These recordings can be:
`standard MPEG Layer I or I1 audio files
`LBR files, converted from Windows WAV format by
`DOS software available from Xing
`MPEG or LBR files encoded by the Transmitter and
`captured by the Capture Tool.
`The server is capable of serving clips (sections of record-
`ings). It is also able to offer different content depending on
`the client’s requested bandwidth. To describe an audio clip,
`Streamworks uses small ASCII text files called ”playfiles”.
`A playfile defines a clip giving the filename of a recording
`and the start- and end times of the clip relative to the begin-
`ning of the recording. This data can depend on the client’s
`requested bandwidth, making it possible to serve clips of
`different quality for each bandwidth.
`To listen to a Xing based Internet radio one needs the
`Streamworks Player (the client software). It can be down-
`loaded freely at Xing’s WWW Site. It supports Windows
`3.x and 95, PowerMacintosh, Solaris, IRIX and Linux.
`
`GMD ’s addition to Streanzworks
`As delivered, Streamworks meets most of the require-
`ments described in chapter 111. When working with Stream-
`works, we found out that Xing’s Capture Tool is not very
`well implemented. It consumes nearly the whole CPU
`power. Furthermore, it has to be re-started for every new
`audio file to create, which leads to unpredictable delays.
`In addition, there is no reflector within the Streamworks
`package. A stream can be either broadcast live or stored for
`later use, not both.
`GMD implemented a reflector and a new recorder for
`Xing’s LBR format audio streams. The software runs on
`Solaris The recorder is programmable and stores the live
`stream coming from Xing’s Transmitter via the reflector
`into audio files according to a desired schedule. It con-
`sumes less than 1% of a Sun SS20 CPU (SuperSparc 2 at
`60MHz). It does not need to be stopped and re-started to
`pioduce a new recording, therefore it can start faster and
`with more accuracy to a time-based schedule.
`With these extensions, Streamworks meets the require-
`ments described in chapter 111. Its LBR audio codec deliv-
`ers acceptable speech quality at datarates below 15kbps. It
`plays sections of recordings with good accuracy. The client
`software supports many platforms, thus reaching a maxi-
`mum amount of listeners. GMD chose Streamworks for
`their Internet “PersonalRC3dio” (see next chapter). Feed-
`back mails show that this service is used worldwide with
`reportedly good audio quality.
`
`B. RealAudio by Progressive Networks
`
`In spring 1996, when we chose Streamworks for our
`projects, RealAudio by Progressive Networks [4] was
`already the market leader for Internet audio streaming, but
`Streamworks provided better audio quality at lower band-
`width.
`With the release of version 3.0 in autumn 1996,
`RealAudio provides several new audio codecs which excel
`those of Streamworks. For radio programmes containing
`mainly speech there is a codec at 6,5kbps delivering better
`audio quality than Streamworks’s LBR at I 1 kbps. When it
`comes to music, jingles etc., the 6,5kbps codec is not usable
`any more, but there are several other low bandwidth codecs
`suitable for music, as well.
`
`The components
`The RealAudio system consists of software only. In
`terms of Fig. 2, digitizer and encoder are parts of a Win-
`dows PC. The digitizer may be any Windows MME com-
`patible sound card (e.g. Soundblaster) while the encoder is
`software by Progressive Networks (“Real Audio Encoder”).
`A fast 486-PC can digitize and encode audio in real time.
`Again, the resulting audio stream is transmitted via a net-
`work
`interface
`to
`the streaming server (“RealAudio
`server”). The RealAudio Encoder is also capable of storing
`the audio data as a file, acting as a reflector and recorder at
`the same time.
`The RealAudio Encoder is Windows software capable of
`encoding audio data directly from the digitizer to a live
`stream and/or a recording. It can also encode an existing
`WAV audio file into a streamable format. A DOS version
`makes it easy to encode the same WAV file into several
`streamable formats with different bandwidths and qualities
`using DOS batch processing. A Solaris CLI version
`encodes WAV and AU format audio files.
`The RealAudio server is available for several platforms
`(Windows NT, Solaris). It can play live streams and sec-
`tions of recordings with good accuracy and chooses
`between files according to the bandwidth of the client’s
`connection. To describe a clip, RealAudio uses text files
`called “metafiles” similar to Xing’s playfiles.
`The RealAudio player can be downloaded freely for
`Windows (3.x/9.5), Macintosh, OS/2 and several UNIX sys-
`tems.
`With its new version 3.0, RealAudio made a great
`progress. It meets the requirements described in chapter 111.
`
`C. Which One to Use?
`
`RealAudio excels Streamworks mainly in terms of audio
`quality/bandwidth relation and player reliability. Not in the
`scope of this paper, but influencing the decision what to use
`are the video extensions. Both systems support video
`streaming, but RealAudio (then called “RealVideo”) works
`much better. We are currently preparing to replace Stream-
`works with RealAudio/RealVideo in our projects.
`
`IEEE Catalog Number: 97TH8280
`
`- ss74 -
`
`ISIE’97 - GuimarSes, Portugal
`
`Juniper Ex. 1043-p. 4
`Juniper v Implicit
`
`
`
`V. GMD’s Interiiet Audio Streaming Services
`
`A. PersonalR @ dio
`PersonnlR@dio [31 is the first German Internet radio
`with both a live stream and seven days audio archives.
`“BSaktuell”, the news channel of Bayerischer Rundfunk
`(BK), Munich, is the first complete radio programme which
`is fed into the Internet in parallel with its terrestrial broad-
`casting and simultaneously stored as digital audio archives
`on the server. This enables a subsequent on-demand access
`to any broadcast of the last seven days via the World-Wide
`Web (WWW).
`
`Currevit Standings
`Perso/zalR@dio uses Xing Streamworks with GMD’s
`extensions described in chapter IV. Its setup (Fig. 3) is sim-
`ilar to the generic setup shown in Fig. 2.
`
`L4
`&
`
`TREAMWORKS PLAYE
`
`TREAMWORKS PLAYE
`
`AI‘ELILITk RECEIVE
`
`I -
`
`I
`
`I
`
`Figure 3: Setup of PersonalR@dio
`
`The program is received by a satellite receiver. The ana-
`log audio signal is encoded by a Streamworks Audio Trans-
`mitter. We use the LBR audio codec at 1 lkbps. This leads
`to an understandable shortwave-like audio quality which is
`sufticient since “BSaktuell” is a news channel and mainly
`consists of spoken words. With 11 kbps, the programme is
`accessible throughout the world as has been confirmed by
`many listeners via e-mail.
`GMD’s recorder software stores recordings every quarter
`of an hour according to the time schedule of “BSaktuell”.
`At the same time, it overwrites recordings which are older
`than a week. This selt-updating archives run automatically
`with no human effort needed.
`The Web interface of PersonalR@dio is driven by CGI
`scripts which always produce up-to-date HTML pages
`when accessed. The user can choose a date and a time out
`of the last seven days and immediately gets the audio
`broadcast of that time. Thus, the user is independent of
`broadcasting times.
`GMD released a similar Internet radio for Mitteldeut-
`scher Rundfunk 121.
`
`Outlook
`For the future, it is planned to provide a database with
`additional textual information. The user can search for top-
`ics instead of broadcasting times and quickly finds the news
`he or she is interested in. To achieve this, GMD plans to
`connect to the broadcaster’s internal digital schedule to get
`content descriptions and timestamps automatically from the
`station. They will be processed in our database giving audio
`clips. Since Streamworks (and RealAudio) is capable of
`playing sections of’ recordings without cutting them physi-
`cally, information will be accessible in the Web a few cec-
`onds after the content is broadcast by the radio station.
`
`B. An Internet Radio using RealAudio
`
`GMD currently does not provide an audio-only set-vice
`im
`using RealAudio, but we implemented “Rundschau
`Internet” (,,Bavarian TV News in the Internet”) [I]. This is
`a German
`language video-on-demand
`service using
`Realvideo, the successor of RealAudio. Hence this paper is
`only concerning audio streaming, we describe a possible
`audio-only service using RealAudio which we could imple-
`ment by using the experience with the system during the
`setup of “Rundschau”.
`
`- _ I1
`
`1
`
`I
`I
`I
`I
`I
`I
`I
`I
`
`Figure 4: Setup of an Internet Radio with RealAudio
`
`The programme is received by a satellite receiver or any
`other kind of radio tuner. The analog audio signal is
`plugged into Line Input of’ the Encoder PC’s sound board.
`The RealAudio Encoder running on the same PC converts
`the digitized audio signal into the RealAudio format. It
`sends the RealAudio stream to the RealAudio server run-
`ning on a Solaris or Windows NT machine. Simultaneously,
`the RealAudio Encoder stores the audio data on hard disk.
`The RealAudio server transmits the live stream or clips
`from hard disk to listening RealAudio players.
`
`IEEE Catalog Number: !)7TH8280
`
`- ss75 -
`
`ISIE’97 - Guimarses, Portugal
`
`Juniper Ex. 1043-p. 5
`Juniper v Implicit
`
`
`
`GMD STREAMING CLIENTS IN THE
`GMD STREAMING CLIEE
`SAME MULTICAST GKOllP
`
`AUDIO SOURCE I
`
`FEEDBACK
`
`-
`
`MULTICAST LIVE STRE
`
`r -I-
`1-
`I
`
`SOUNDBLASTER
`
`/
`
`
`
`6
`
`HARD DISK@)
`
`REALTIM
`
`ENCODER
`
`I& GMD RECORDER
`I
`I
`
`I I
`PC(LINUX, WINNT), SUN(SOLARIS)
`PC(LINUX, w95)
`L - - A L - - - - - J
`Figure 5: Setup of the next-generation streaming
`system
`client's feedback. The first results are promising and the
`combination of PAC audio and RTP/UDP/IP multicast
`serves well for live services.
`Our first clients (for SUN/Solaris, SGIARIX, InteVLinux
`and Intel/Windows 95) support PCM, PAC and partially
`MPEG Layer 3, although the server is format-independent.
`
`VI. The Next Generation
`The discussed commercial streaming system have the
`following drawbacks:
`1. More than one request to access the same live stream
`results in a unicast transmission for each request. There
`is n o support for IP multicasting to reduce the network
`load.
`2. No feedback information from the client is used to
`dynamically
`.adapt an outgoing stream to network
`conditions.
`3. No dynamic creation / provision of multiple service
`qualities (e.g. encoding / datarates) is provided from of
`a single archives file.
`This led to the decision to design and implement our own
`streaming system.
`This chapter discusses components of the next genera-
`tion of streaming system currently being developed by
`GMD .
`A. The Coinponents
`The streaming system includes the following compo-
`nents (see also Fig. 2):
`The t d t i n i e audio encoder which supports several
`codecs.
`The streaming server according to Fig 2. is codec-inde-
`pendent and consists of three parts:
`A request handler which handles the clients connection
`requests by starting the appropriate streamer for each
`request, or by referring
`the client to an existing
`multicast group.
`An ori denzand streamer providing on-demand-streams
`to
`the clients. It adapts the bitrate dynamically
`according to the clients feedback.
`A live strsanier which provides a live-stream to a
`unicast address or a multicast group. Bitrate adaptation
`for unicast streams is mandatory. Multicast streams
`implement a bitrate adaptation via RTCP.
`The reflector provides the encoded audio stream to both
`the live streamer and the recorder.
`The recorder stores the encoded streams from the
`encoder and probably additional information on a hard
`disk.
`The streanzing c h i t requests, receives and plays the
`audio stream.
`The following two components may become necessary in
`the future:
`A tmnscoder allows transcoding of available streams
`according to the client requirements (e.g. transmission rate,
`presentation format, e.g. on-the-fly transcoding from PAC
`to MPEG Layer 3).
`An enhanced on-demand-streamerhequest handler com-
`bination, allowing the server to bundle more than one client
`request to the same stream within a given period (lets say
`10 seconds) to one multicast group.
`B. Inzplemetitation Status
`GMD's current implementation of the above presented
`streaming system consists of the streaming client, the live
`streamer and an encoder supporting PAC and MPEG Layer
`3 streaming and dynamic rate adaption depending on the
`
`VII. References
`
`[ I ] Bavarian TV News on the Internet
`WWW page at /itt~~://01--tG~~iizc/.e/e
`[2] MDR info
`WWW hompage at ~ ~ ~ / ~ : / / J ~ Z ~ / f ~ j f ~ f ~ / . ~ / ~ / i i f ~ f J ~ - ~ f ~ ~ ~ ~ / / ~ ~ J i Z / /
`[ 3 ] PersonalR@dio
`WWW hompage a( ktt]J://bf:~~rizcl.cl~
`[4] RealAudio by Progressive Networks
`WWW hoinpage ai http://wwv, 1-euluudio.coin
`IS] Internet Protcol
`Internet Network Working Group, Request For Comment 760,
`January 1980
`[6] Transmission Control Protocol
`Internet Network Working Group, Request For Comment 761.
`January 1980
`[7] J. Postel: User Datagram Protocol
`Internet Network Working Group, Request For Comment 768 ,
`August I980
`[8] RTf: A Trun.yior? Pmtoc.o/Jor R e t r l - T ~ ~ ~ e Applic.utiom
`IETF Network Working Group, Audio-Video Transpolt Working
`Group: Request for Comments: 1889, January 1996
`[9] W. Stallings: The Hypertext Transfer Protocol
`connexions Vol 10 No 8, August 1996
`[IO] Xing Streamworks at htt/'~//~vvww.n.iiljc~c/z.
`('om
`
`IEEE Catalog Number: 97TH8280
`
`- SS76 -
`
`ISIE'97 - GuimarLs, Portugal
`
`Juniper Ex. 1043-p. 6
`Juniper v Implicit
`
`