Presented at the 1996 NAB Broadcast Engineering Conference
Steve Church
Telos Systems
Cleveland, Ohio USA
EUonline use by Permission
The Internet has the potential to offer radio broadcasters the ability to extend the reach of their existing programming to the world and the possibility to begin new programming services.
While offering new avenues for existing broadcasters, the net also poses a threat - as it allows an unlimited number of new audio services to be initiated at relatively low cost.
I suspect that most of you who are in the radio broadcasting community will, before too long, be involved in Internet audio broadcasting. This paper offers an overview of where the technology is today, along with some discussion and predictions for what might be coming. It also has a description of systems which may be used for netcasting today. Consideration will be given to some important limitations of the Internet which would impact broadcast-like audio projects, and possible solutions.
As both the Internet and the systems to support audio transmission are emerging and fast-changing technologies, there are many novel questions.
For Broadcasters: Net Profit or Net Loss?
Byte Magazine, in the February issue, had a cover story entitled "How the Internet will Replace Radio Broadcasting." I've been in the broadcasting industry my entire working life, so I have mixed feelings about this prophesy: That"s why I have entitled this paper "The Internet and Radio Broadcasters." Either way, this is something that will happen to us, and we better understand it. That is the first goal here. The second is to introduce a new technology for those sufficiently brave to start an Internet broadcasting service.
I believe net radio follows two important trends:
It is clear that the transition from analog to digital has been ongoing and thorough. While the transition to packet digital is less clear today, it is equally inevitable. Both are caused by the economics of computing. When the demands of general computing made digital processing power available at a low price-point, the transition of audio to digital was initiated. The same process will drive the move to packet transmission of audio. Computer networking is exploding at all levels: LANs and WANs are on a speed and quantity growth curve even faster than that of processing power 10 years ago.
George Gilder, Peter Drucker, Alvin Toffler and other observers have noted that a key element of our "post-modern" times is the ongoing demassification of media, and its replacement by forms less centralized and hierarchical, more individualized and interactive. Only a few years ago, most radio listeners and television watchers were served from a very small number of programming sources: Four television networks have become 40 on cable and now over 200 on the DSS satellite. In the computer world, we had four or five major content sources. We now have over 500,000 Web sites - a transition that took place over only about two years.
This net prophet (sorry!) is about 97% convinced that radio-like audio services over the Internet will grow to become an important supplement to current broadcasting. What does this mean for you? If the company you work for sees itself primarily as an owner of RF channels, the net is not going to be your friend; on the other hand, if your company"s mission is seen as "producing appealing audio programs," it and you may find the net ultimately to be a liberating and lucrative new path to listeners, albeit a frustrating one at the outset.
This is not something happening in some distant future. The selection of audio now on the net is surprisingly varied, given that the supporting technologies are only about a year old. You can listen to live and recorded news and sports from the big networks like ABC, ESPN, and NPR. You can listen to any of a few dozen radio stations that have taken their broadcasts to the net. You can sample music from major record companies as well as unsigned fledgling bands.
Apple, CompuServe, Hotwired and many others are providing ongoing audio-on-the-net programming.
Existing broadcasters have the advantage over new programming entrants that they already have expertise and personnel in place to create audio for the net. It is possible to begin modestly by simulcasting fare presently being produced for the air.
The Challenge: Bandwidth
Pumping high-fidelity audio over the Internet is not easy. CD-quality sound consists of
two 16 bit samples, 44.1 thousand samples per second, for a bitrate of over 1,400 Kbps.
With modems maxing out at 28.8 Kbps, you can see we have a problem. Even with the most
advanced technology envisioned by modem designers, speeds will top in the 30 to 40 Kbps
range. Solutions, fortunately, are at hand. We will look to two enablers: 1) Faster
connection methods, and 2) Compression techniques to reduce the number of bits.
Boosting The Bits: Faster Connections
The current generation of modems, at 28.8 Kbps, are fast enough for low-fidelity audio, but are not going to offer your listeners anything like the audio experience they"ve come to expect from FM-quality broadcasts. With the right compression technology though, they could link listeners to something like AM grade audio for voice services.
The first step up from 28.8 modems is ISDN. This service is increasingly being used by net-surfers who crave speed, and the phone companies are both making this service more available and making the cost for residential use more reasonable. At minimum, this service provides 56 Kbps, with rates to 128 Kbps possible. ISDN is widely available now, generally at around a 20% premium over traditional analog lines. Difficulty ordering and setting-up ISDN and lack of low-cost interface equipment has slowed widespread adoption, but these problems are being solved. It is now common to see ISDN "modems" on the shelves of computer stores next to the analog units.
ISDN Basic Rate Interface lines in the USA. The number in service will top two million by 1999, according to Dataquest estimates.
A faster version of ISDN, called Primary Rate, and T1 telephone circuits are widely used by businesses and universities for state-of-the-art net connections at 1,500 Kbps.
But the quest for bandwidth does not stop there. ATM and ADSL are new technologies which offer bitrates from 1,500 to over 600,000 Kbps. While ATM may be useful in some future scenarios, it is much too expensive for home use now. ADSL, though, is thought of by the Telcos as a way to compete with cable companies while using their existing copper wires, so this service may someday offer high bandwidth at low cost.
Very recently, two telephone companies have announced trials of ADSL. U S West and GTE said they want to provide ADSL to home users at 6,144 Kbps as a faster alternative to ISDN. U S West says they expect pricing to be in the $35-40 per month range.
Cable companies, not wanting the phone companies to prevail in the race to provide high-speed home Internet service, have plans afoot to piggyback high-speed net connections over their fat-capacity coax cables. AT&T, Intel, Hybrid Networks, Hewlett-Packard, Motorola and others sell cable modems for a few hundred bucks, and a few major cable outfits have announced trials of this technology. Speeds vary from 500 Kbps to 30,000 Kbps.
Time Warner has been running a cable modem test called Linerunner in Elmira, N.Y., since last July. Although it currently has just 200 participants, Time Warner has ordered about 100,000 cable modems from Motorola and Toshiba. Time Warner's plan is to offer similar Internet access services in Akron, Ohio, and San Diego by next fall.
Additionally, Linerunner is working with Time Warner's publishing arm to offer local publication content and Time Warner material currently on the Internet, like its popular Pathfinder Web site.
TCI, probably the country"s largest cable operator, has trials planned to be underway soon in the San Francisco area as part of a major initiative called the @home project.
Cable could become a high bandwidth path into the Internet - and to your programming.
Compression: Squeezing the Bits
Whatever bandwidth we have, we will need to optimize the audio bitstream to get it to fit with maximum fidelity.
Most broadcasters are familiar with the codecs which allow stereo CD-quality audio to be sent along 128 Kbps ISDN telephone channels, a reduction of about 12:1, so we"re not too far away from something that could be used for FM-like quality. If we could squeeze maybe two times as much, we could get a result that could be carried by one ISDN channel, or by a smallish piece of an university or corporate T-1 connection.
Fortunately, this is now possible. The latest research into perceptual coding has resulted in methods which are able to operate beyond 25:1, achieving AM quality for listeners with 28.8 modems and decent Internet conditions. With ISDN-grade connections and the latest compression methods, FM-stereo quality audio is possible today.
Net Problems
So far, we have considered only the link to the user in our quest for fast bitpipes, assuming that the infrastructure of the net itself will reliably handle the intermediate connection. As users of the World-Wide-Web know, this is not generally the situation we currently have. The tremendous growth of net users and services over the past couple of years have put a strain on its capacity to carry that backbone and local providers are constantly addressing with more and faster connection paths and routing equipment. Our assumption is that providers will rise to the challenge and will build the capacity that users demand, as they say they will.
MCI, for instance, has recently upgraded their portion of the net backbone from 45 Mbps T3s to multiple 155 Mbps links.
A specific problem which has to be considered in any practical real-time audio delivery
is the limitation of the net"s TCP/IP protocol. This is the low-level near-universal
language of net communication, specifying how data bits get accumulated into packets and
routed from the source to the destination. TCP guarantees reliable end-to-end delivery,
but at a cost: Internet routers are allowed to drop packets when they are above capacity,
and the re-transmission of dropped packets can cause long delays when the receiver sends a
request for lost data and then has to wait for it to be re-sent. New protocols have been
devised and are beginning to be implemented which offer connections optimized for
real-time transmission.
The Promise of "IP Multicast"
Yet another problem standing between the net-as-it-is and a useful broadcasting system is that, normally, each listener requires a unique path from his receiver to your source audio. This is tremendously wasteful of bandwidth and is very expensive, because a provider of audio must support a net connection at his end which can carry all of the bits individually for all of the active listeners! Without attention to this problem, it is not economically feasible to support more than a few hundreds of listeners. What we need is some method for the network itself to replicate the streams as close as possible to each listener, a process called multicasting.
Multicasting cannot use TCP/IP because this method requires a bi-directional path from the server to the user in order to support re-transmission of lost packets. A method called RTP or Real-Time Protocol is proposed and has recently been endorsed by Netscape and others.
An important component of any multicasting system is the new RSVP or Resource Reservation Protocol. The present situation is that no bandwidth is ever guaranteed to a user. This is not too much of a problem for text and graphics, but real-time media must have reliable delivery of packets. The RSVP scheme is proposed as a way to communicate to the Internet routers that a certain bandwidth is required by a user. The net then either grants the request or reports to the user that the request cannot be fulfilled. RSVP is not yet implemented, but there are predictions that we could begin to see it within a couple of years.
Multicasting is now at an experimental stage, with a few thousand wired Unix-heads trying the technology in an application called MBone (Multicasting Backbone). Observers have said that the Mbone is about where the Internet was about 10 years ago - and that it may be ready for take-off.
It appears that the underlying MBone technology is able to provide what we need to have a real broadcast-like service with a capability to serve large numbers of simultaneous listeners.
Another possibility for the future involves a completely new approach called ATM or Asynchronous Transfer Mode. This replacement for the current IP system is well-suited to real-time multimedia as it naturally guarantees connection bandwidth. Some proponents predict that ATM will eventually replace IP, but it is unclear to me whether this technology will prevail.
Whichever way this goes, there is growing pressure on the network infrastructure to support real-time audio and video, and some method of ensuring reliable bandwidth will ultimately be adopted.
What Can You Do Now?
So, is it possible to "netcast" now? The answer is, as you might now expect, well... yes, and no.
To reach the vast number of users with 28.8 modems, something like AM to about a few-hundred listeners is doable - and a few intrepid pioneers are now paving the way.
Smaller numbers of users, those who have higher-speed connections, can get FM-quality stereo audio now, again with the limit for number of active listeners topping out at a few- tens or hundreds.
It is also possible to offer audio on a non-real-time basis to almost any net user. In this case, high-quality is possible regardless of connection speed or network conditions because the audio is assembled and stored on the user"s hard disk before playback.
The Internet was not designed for multimedia, and the network is not at all ideal for real-time audio feeds. Nevertheless, manufacturers have devised clever technologies to overcome the net"s limitations. Each have their own approach, but share in common the two essential elements:
Some of the providers of equipment and expertise are listed below.
The most widely used system (a claimed over 4
million players downloaded), provides very low fidelity audio at a bitrate that even 14.4
modems can communicate. Version 2.0 works with 28.8 Kbps modems or better connections and
offers AM quality. Uses a proprietary method to deliver the audio packets, which allows
audio to continue to flow (with lower quality) even with packet loss.
Provides both low-fidelity/bitrate and high-fidelity/bitrate transmission
technology. Lo-fi uses proprietary scheme, while hi-fi uses MPEG Layer II. Offers higher
audio quality than RealAudio when the connection can support it. Also can provide a
companion MPEG video system.
Focuses on high-fidelity, broadcast-like applications. Advanced perceptual encoding (enhanced MPEG Audio LIII) is performed in special-purpose DSP hardware rather than in a general-purpose PC, raising cost but improving fidelity. Uses HTTP for packet transfer, which allows audio to be distributed by modified Web servers and has the advantage that it can traverse most firewalls. Expects to migrate to RTP and IP Multicast when practical. Provides highest available audio quality on the net today.
The Cost Issue
Until we get multicast capability within the Internet and the audio software to support it, we have the problem that we have to support each listener all the way to the server. This bandwidth is expensive. To support 1,800 simultaneous listeners for a mono AM-like signal requires a T3-grade Internet connection at a cost of about 50K/month. To support FM-stereo quality, twice the bandwidth per listener will be required.
Clearly, this does not work with the existing radio broadcast economic model, as the cost to support a listener is greater than the advertising revenue likely to be generated. Compared to the Internet, the cost for a traditional broadcaster to reach each listener via RF transmission is very low.
Of course, there may be some audio content targeted to niche audiences which could be economically viable. Specialized sports, business information, foreign language programming, etc.
Web advertising (on the associated graphics pages) is another route to revenue. While AQH on the net is limited, the cume could be very large - and this is what matters to Web advertisers. While Web advertising is a dynamic field and not too much is yet known about it which is universally accepted, it does not seem that number of impressions is as important as in radio advertising. Web advertising is generally considered to be more effective in terms of response per impression owing to the targeted nature of Web users and the ability to act on a message immediately by clicking on the link.
Another solution to the economic problem is the introduction of a method to automatically charge net users for their listening. Many efforts are underway to implement secure and ubiquitous user payments. An ideal solution would be one which permits billing in "micro" amounts efficiently so it would be possible to have an hourly or per-minute listening fee in amounts small enough not to be a barrier to attracting listeners, but would nevertheless afford a revenue stream sufficiently large to make netcasting a viable business.
Music Licensing
One concern regarding the feasibility of netcasting is in the area of music rights and licensing. Recordings are usually protected by two kinds of right: one is the familiar BMI/ASCAP composer copyright; the other is the mechanical or performance right, which is most often held by the recording company.
With regard to BMI and ASCAP, BMI wants 2.1% of all Internet only revenue generated directly from an Internet site. Rebroadcasts of spots from a re-translated licensed radio station do not count. There is a base charge of $500 per year if there is no revenue or if revenue is less than $500. ASCAP is similar: 1.615% of revenue or $500 flat on the first $500 direct revenue.
With regard to the performance right, it appears that streaming on the Internet is permitted with no additional payment. The new Digital Sound Recording Law addresses this possibility,
Quote from section 114(d)(1)(C)(iii): (page 369):
"To simplify licensing practices, (law) provides a "through the listener" exemption intended to permit transmitters including cable systems, direct broadcast satellite service providers, and other multichannel video programming distributors, simultaneously to retransmit to the listener non-interactive music programming provided by a licensed source. To qualify for this exemption, the retransmission must be simultaneous with the transmission and authorized by the transmitter; and the original transmission must be licensed by the copyright owner of the sound recording. For purposes of this exemption, retransmissions are deemed to be "simultaneous" even if there is some momentary time delay resulting from the technology used for transmission or retransmission. Thus, (law) exempts retransmissions from liability for copyright infringement where a noninteractive music programmer transmitter has obtained a public performance copyright license from the copyright owner of the sound recording."
The interpretation of record companies we have contacted is that in the case of a licensed over-the-air radio station doing netcasting, there are no additional copyright fees owed to the record company. In the case of establishing an "Internet only" radio station streaming in real time, we presume the same holds true, but this is less clear.
However, here is an important warning: The record companies will very probably not take kindly to "interactive" systems which permit selective accessing of individual recordings. Understandably.
Details of the Telos Internet Audio Suite
The Telos Internet Audio Suite comprises an evolving range of hardware and software solutions.
The audio coding method is the most powerful possible, an optimized MPEG LIII perceptual technique. This is a variant of the method widely used in the broadcasting industry for transmission of high-fidelity audio on ISDN telephone lines. It has been developed over the course of a decade by people dedicated to professional audio coding.
In all international listening tests conducted under the auspices of ISO, MPEG LIII impressively proved its superior performance, maintaining the original sound quality at a data reduction of 1:12 (60 Kbps per audio channel). With the recent optimizations, mono 8 kHz is possible at bitrates down to 16 Kbps and FM-stereo grade quality is attainable at 56 Kbps.
Thus, 28.8 Kbps modem users can listen with reasonable quality, while those with higher speed ISDN, university, or corporate network connections enjoy full broadcast quality audio.
The transmission method is via the usual Web HTTP file transfer over standard TCP/IP.
This has the advantage that any firewall which permits normal Web browsing will allow the
audio to pass.
What follows is a rudimentary description of what would be required to implement an Internet audio broadcasting system using the Telos MPEG LIII technology.
The system serves multiple users simultaneously. It is capable of various bitrates to
handle the differing delivery pipe speeds. Either stored files or real-time audio can be
served on a streaming basis.
The software-only coder converts PCM audio files in Apple AIFF, Unix .au, or Microsoft .wav format and stores them for subsequent downloading or streaming. Versions are available for PCs running MS-DOS, Sun workstations running Solaris, PCs running Linux, and for machines running NeXTSTEP 3.
Execution time depends upon the computer being used. A one minute audio segment takes 14 minutes on a 486/66 or 5 minutes on a Sun Sparc 10.
Bitrates from 8 Kbps to 320 Kbps are supported.
The Telos MPEG LIII approach gets much of its power from not having the constraint that real-time encoding must be done in a general-purpose PC platform. Our streaming encoder is a special purpose hardware box which harnesses the processing power of five DSPs.
In the prototype, audio is input via two pro-audio grade balanced XLR connections. Output is taken via an RS-232 or an X.21/RS-485 port. Production units will add an AES/EBU input and TCP/IP output options.
The unit is housed in a single-rack-unit high enclosure. Metering is provided on the front panel to set audio input level. Bitrate and other mode settings are made via a dedicated RS-232 port in the prototype. Production units will permit remote operation via a TCP/IP Telnet connection.
Each encoder supports a single bitrate. Multiple units may be used to serve streams to
users at various rates.
The server duplicates the input bitstream in order to serve multiple simultaneous listeners. The present implementation uses Pentium PC running a modified Web server under the Unix variant Linux. This configuration is expected to be able to support as many as 100-200 real-time connections. Multiple boxes or higher-performance machines such as Sun Sparc stations or Silicon Graphics computers could serve larger numbers of listeners.
In the prototype set-up, encoders interface to the server via an RS-232 connection.
Production units are expected to use TCP/IP with a 10baseT Ethernet interface as the
transport medium.
Client software is available for Windows platforms. At least a 486/66 machine is required for real-time play. On a Pentium 90, approximately 33% of the processor"s power is consumed, permitting a user to browse the Web, do word processing, etc. while the audio decoding and play process continues in the background. The audio is delivered to the computer"s Soundblaster compatible audio card.
A client for Power Macintosh computers is nearing completion. It is expected that processor load will be approximately the same as with Pentium computers.
Unix clients are under development.
All decoders permit either playback from a local file or from a network connection. The user interface is much like a CD player, with stop and play buttons, status indicators, and a "track timer" display.
The decoding process may be initiated by simply clicking on a link using the Web browser in the usual "helper app" fashion. Of course, the browser's MIME type preferences must be set to match the player"s. The player may also be used stand-alone by entering a file name or URL directly.
A hardware DSP-based decoder is available which plugs into a PCs parallel printer port
for users who need to retain the full power of the main processor.
Conclusion
The Internet is not going to go away and it will become a medium for the distribution of audio and video content. There are over 20 million people around the world with Internet access, a number which is increasing rapidly. Approximately 35% of Americans have a PC at home; among families with incomes over $50K, home computer penetration approaches 60%.
The Economist magazine said this recently about the impact of the microchip on media:
The first computer revolution was about number and word crunching; another one is now underway - a revolution in communication. A machine which transforms communication impinges far more radically on people's lives than one which transforms computation. Why mass media when information can be consumed in individualized packets. The next limiting factor is not going to be the ability to imagine the future, or to invent it. It is the willingness to embrace it which will matter most.
Inventors and engineers are closing in on delivering useful technology for netcasting.
It will soon be up to you: the suppliers of capital, the creators of programming, and the
providers of broadcast engineering expertise to bring it home.
Steve Church may be reached by email at steve@zephyr.com, or by phone: +1 216 241.7225.
Telos, Telos Systems, and Zephyr are property of TLS Corp. All other trademarks are the property of their respective owners.EUonline presents commentary and opinion to foster informed discussion and reasoned debate. The views of our commentators do not necessarily reflect the views of EUonline and National Public Radio.
