Data conferencing might be the first "killer" multimedia application
Andrew W. Davis
A revolution is quietly changing the desktop computing landscape. The resolution of DSP (digital signal processor) standards and the development of programming interfaces promise to dramatically change the nature of the PC peripheral board business. In a short time, companies marketing dedicated modem, sound, codec, and similar add-in cards will replace those dedicated cards with single-card multifunction products. What is making this feasible are new (RAM-based) DSP chips and the multitasking software that enables a single board to take on multiple personalities: You simply download the appropriate algorithms for communications, speech processing, sound editing, or even image/video decompression (see the text box "The High
Cost of Videoconferencing?").
While business users have been slow to embrace desktop multimedia, this new breed of mixed-media modems is appealing because it holds the potential to save time and money as it changes for the better the way many of us do our work. The mixed-media modem can bring real-time, remote, interactive sharing of data to the desktop in a low-cost, easy-to-use platform that takes advantage not only of universal POTS (plain old telephone system) lines but of newer LAN and WAN (wide-area network) connectivity as well.
Companies are embracing mixed-media audiographics, or data conferencing, as an enabling technology for a variety of next-generation applications. These include lower-cost, higher-performance customer technical-support centers, remote presentations and sales calls, distance learning, telecommuting, and a variety of other remote collaboration applications. Whether the business driver is a need to reduce office space, to comply with the Americans with Disabilities Ac
t, to meet local air pollution/traffic-control regulations, to enhance the effectiveness of remote collaboration, or to improve employee quality of life with work-at-home programs, desktop data conferencing is an increasingly viable option. Indeed, data conferencing might be the first "killer" commercial multimedia application.
Teleconferencing
Early teleconferencing products were essentially sophisticated telephones that provided a convenient way for one group of people in a room to converse with another group at the other end of the phone line. The current generation of these products relies on DSP technology to provide high-fidelity audio by balancing multiple microphones and providing signal processing that effectively eliminates room echoes and line delays.
Videoconferencing--which adds video of the meeting rooms to the transmission--allows distant colleagues to meet without the expense, waste, and inconvenience of traveling. Because videoconferencing requires significant investments in
equipment and often entails the use of dedicated facilities with special communications lines, its applicability and appeal have been limited.
While traditional videoconferencing systems enable groups to see and hear each other, these systems are not optimized to share on-line information such as spreadsheets, word processing documents, presentation files, databases, scanned or acquired images, and so on. Unfortunately, much of today's business information is on-line--generated and stored on PCs and other desktop workstations. Recognizing this disconnection, several vendors of dedicated videoconferencing equipment are incorporating the data elements into their desktop systems, even though most of these systems today suffer from video compression bottlenecks and requirements for relatively expensive, high-speed communications interfaces (see the text box "DSPs and the PC Mainstream"). While these limitations rule out truly interactive, broadcast-quality desktop videoconferencing on PCs using low-cost mu
ltimedia peripherals, what's practical today are data conferencing, collaborative computing, and individual-to-individual communications over a standard phone line (see the table "Teleconferencing Options").
Document Conferencing
Document conferencing is a new variant of screen sharing (previously called remote log-in). Instead of having full control of a local system, a remote user shares one or more designated windows with a local user. Most document-conferencing solutions use a whiteboard model. After establishing a desktop-to-desktop connection, the session presents a whiteboard window that both participants view and manipulate.
The whiteboard software includes a complement of drawing, painting, and annotation tools for brainstorming and sketching, enabling you to emulate the interactive discussions that often take place in real meeting rooms. Bit maps of the whiteboard are JPEG-compressed prior to being sent over the wire, and you can save snapshots of the whiteboard for future reference
. However, if you want to make existing information from a spreadsheet or other file available, you must "cut" from the application and "paste" the bit map onto the common whiteboard. The whiteboard data is static--changes made in the collaborative window do not affect and are not affected by the "real" data.
A variation on this theme involves pasting objects--rather than bit maps--onto the whiteboard. Using Windows' OLE features, an OLE-compliant application such as a spreadsheet can provide server services to the collaborative program acting as a client. Embedded objects can be edited in the server application, and the editing changes are reflected in the client (whiteboard) application, after an Update command has been executed.
The next step up in document-conferencing capabilities is interactive file or application sharing. Here, you select a window or file that a remote partner can access, and then both of you can review and modify the document through the shared session. The main advantag
e is that you can directly modify source data, not an intermediate bit map. Most of the vendors marketing whiteboard products intend to add file sharing to their offerings.
Various technical approaches can be used to enable application sharing. One approach captures the Windows GDI (Graphical Device Interface) screen drawing commands and sends these commands over the phone line to the receiving system, where they are executed in parallel. Sending drawing commands and not bit-mapped images makes efficient use of the communications bandwidth while granting the remote and local users access to the same screen.
With document conferencing, you can send text, images, graphics, spreadsheets, and drawings over standard telephone lines using common modems, but voice is not part of the transaction. Because the data is digital, document conferencing readily lends itself to any computer-compatible communications channel, including LANs and WANs. Many of the document-conferencing solutions available today fo
r POTS and LANs require the use of a separate phone line for simultaneous voice and data conferencing. This obviously doesn't work if access is limited to a single phone line or if you are working from a laptop on the road.
Data Conferencing
Data conferencing adds a key element to the document-conferencing equation: simultaneous voice and data transmission on the same communications line. Data-conferencing solutions digitize voice and treat it as one more element in the data stream. Many data-conferencing products were developed and marketed as subsets of desktop-based videoconferencing solutions.
While the host PC can readily handle document conferencing, voice coding for real-time interactive communications over a phone line requires the greater processing power DSPs provide. Fortunately, the new breed of mixed-media modems can accommodate voice coding as one more element in the data management library, although performance may be affected.
Voice is typically digitized at a 12-bit re
solution (72-decibel dynamic range) at 8000 samples per second; this is sufficient to handle the approximately 3500-Hz bandwidth of most human speech. A simple "companding" system encodes each speech sample into an 8-bit value, producing a 64-Kbps data stream that exceeds the bandwidth of most modems and phone lines. This results in the need for speech compression, and numerous algorithms have been developed, reflecting a wide range of performance, voice quality, and cost trade-offs. For data-conferencing applications, the goal is to code an analog speech signal into compressed digital format, transmit the data, and then decode to an analog waveform in real time. For personal conferencing using POTS, transmission bandwidth is the primary obstacle.
Two types of voice coding technology exist today. The first type is waveform coders, which deal with signals on a sample-by-sample basis, using only the output signal in the coding process. Waveform coders such as ADPCM (adaptive differential pulse code modul
ation) make no assumption about the source of the input signal. Computationally simple ADPCM techniques produce bit streams with unacceptable voice quality at data rates below 24 Kbps and with unintelligible speech at rates below 16 Kbps, making them unsuitable for widespread POTS-based telecommunications. Source or parametric coders (or vocoders) encode speech signals in terms of parameters that drive a speech production model based on human vocal tract shape and excitation levels. Vocoders are computationally demanding but can operate at much lower bit rates than waveform coders (see the table "Telecommunications and Speech Coders"). Some speech-coding algorithms combine various techniques.
The most common vocoder technology in use today is CELP, or code excited linear prediction, which uses "codebooks" to quantize the input signal. Basic sounds are stored in the codebook and are then modified in amplitude and pitch to reproduce--in a fashion consistent with human vocal-cord anatomy and function--the
input voice. The transmitter decomposes sounds into their codebook values and sends pointers plus modifying parameters to enable the receiver to reconstruct the source voice.
The CELP algorithm provides excellent speech quality and is well suited for teleconferencing, although some listeners complain that the resulting voices sound somewhat artificial. CELP is compute-intensive, however, straining even DSP resources. On a 16.7-MIPS DSP, 4800-bps CELP compression can consume approximately 90 percent of the DSP processing resources. Variations on CELP take advantage of higher output bit streams (less compression) or enhanced codebook search algorithms to reduce the computational load. One recent modification is CELP+, developed by Bell Labs and used in AT&T's PC-based TeleMedia conferencing system. CELP+ produces high voice quality at 6400-bps bit streams and consumes fewer DSP MIPS than earlier CELP approaches. The CELP+ algorithm and the V.32terbo modem algorithm can execute on a single DSP3210-based
communications subsystem. Coded speech data is multiplexed with other digital data within the overall transmission envelope provided by the modem (19.2 Kbps), resulting in a POTS-compatible, desktop audiographics engine.
Performance Considerations
Data conferencing applications can place strenuous demands on desktop computers. Analysis of the MIPS requirements for different DSP tasks shows that the combination of high-speed modem services and concurrent speech compression consumes nearly all of today's mixed-media modem DSP (16.7-MIPS) resources (see the figure "DSP Resource Allocation"). Attempting such a task without a DSP, even on the new Pentium and PowerPC processors, won't provide satisfactory performance.
Once the connection is made, all data must fit into the performance envelope of the modem; this is where modem speed really counts. V.32 is the practical bottom end for data conferencing, and the new V.34 modems with 28,800-bps capabilities promise to make personal teleconferencing ev
en easier and more interactive. Depending on the compression algorithm executed and the throughput of the modem, voice will consume 30 percent to 60 percent of the modem bandwidth.
Data-conferencing applications present multidimensional challenges to hardware and software designers. Design trade-offs involve compression factor, speech quality, computational complexity, and line delay. Compression algorithms are bounded on the high end by available MIPS and on the low end by the output bit stream, which must not exceed transmission-line bandwidth. More sophisticated modem algorithms like V.34 consume more MIPS but support more bandwidth, which, in turn, accommodates less compressed, more realistic speech.
VoiceSpan and VoiceView
An alternative technique to incorporating voice in PC-based personal conferencing is now available through VoiceSpan technology from the AT&T Bell Laboratories group. VoiceSpan increases the capacity of existing phone lines by splitting a single line into three virtual
channels: one for voice or low-quality audio, one for data, and a third virtual channel for control information. With VoiceSpan, users can talk over the phone while simultaneously sending data, faxes, still images, or data from file-sharing applications.
Unlike other products, VoiceSpan is not based on speech coding or analog signal compression. In fact, voice is not converted to bits-per-second data. Rather, VoiceSpan channel coding is based on an extension to the digital data techniques employed in full-duplex, equalized, echo-canceled modems that provide simultaneous communication of both digital and analog information. A DSP digitizes and maps the analog voice signal into a combined analog-and-data signal suitable for communication through a POTS line over modems using QAM (quadrature amplitude modulation) techniques.
VoiceSpan also defines methods for automatically or manually originating and answering calls, as well as for interoperating with standard phones, faxes, and modems. The first
product to embed VoiceSpan technology is the Paradyne DataPort 2001 modem. The 2001 provides 14.4-Kbps transmission when acting as a standard modem, or 4800 bps for data while simultaneously transmitting voice. AT&T Paradyne bundles the product with FarSite whiteboard software from DataBeam, and the technology has been licensed by various companies in the phone, games, computer, fax, and copier businesses.
Early this year, Radish Communications Systems announced a voice-modem technology, called VoiceView, that enables the integration of voice, data, and fax over a POTS line within a single phone call. The mixed-media communications are not simultaneous, however; you switch between them sequentially.
Future
For professionals working on PCs and workstations who need to collaborate with others at different locations, the new DSP-based multimedia peripherals could become an important computing companion. Combining collaborative data sharing with simultaneous voice and operating over standard phon
e lines, data-conferencing solutions will let globally dispersed users with phone access work as if they were in their offices and at their desks.
Teleconferencing Options
Users can match features to operational requirements and communications links.
DEDICATED
STANDARD AUDIO VIDEO-
PHONE/FAX CONFERENCING AUDIOGRAPHICS CONFERENCING
Voice Yes Yes No Yes
Fax/modem Yes/no No/no No/no No/no
Whiteboard/file sharing No/no No/no Yes/no Yes/no
Video No No No Yes
PC/stand-alone Stand-alone Stand-alone Stand-alone Stand-alone
Suitable for POTS Yes Yes Yes No
Conference focus Two- Group-to-
Group-to- Group-to-
way group group group
DOCUMENT PC DATA PC VIDEO-
CONFERENCING CONFERENCING CONFERENCING
Voice No Yes Yes
Fax/modem Yes/yes Yes/yes Yes/yes
Whiteboard/file sharing Yes/some Yes/some Yes/some
Video Still image Still image Still image
PC/stand-alone PC PC PC
Suitable for POTS Yes Yes No
Conference focus Individuals Individuals Individuals
Telecommunications And Speech Coders
BIT STREAM DSP REQUIREMENTS
ALGORITHM (KBPS) (MIPS)
Parametric coders
LPC10 (FS1015) 2.4 9 to 11
CELP (FS1016) 4.8, 7.2
15
CELP+ 6.4, 6.8 5.8 to 6.3 (AT&T)
VCELP (cellular phones) 7.95 15 to 19
LMCELP (low-memory CELP) 4.8, 7.4 13 to 15
LDCELP (low-delay CELP) 8 19.6
G.728 12.8, 14.4, 16 16.5
TrueSpeech 6.8, 7.2, 8, 11 7 (DSP group)
Waveform coders
G.726 16, 24, 32, 40 5 to 8
G.722 48, 56, 64 9 to 10.2
G.711 48, 56, 64 0.6
Modems
V.22bis 2.4 5
V.32bis 14.4 9.6 to 9.9 (AT&T)
V.22terbo 19.2 10
CCITT G3 fax 9.6 9.5
Data courtesy of DSP Software Engineering, Inc. (Bedford, MA)
Figure: DSP Resource Allocation
Typical demands on modem bandwidth during audiographics teleconferencing sessions. A combination of
high-speed modem services and speech compression can consume nearly all of a typical DSP's resources (16.7 MIPS).
Andrew W. Davis is an independent marketing consultant in Southborough, Massachusetts, focusing on high-technology business development and marketing communications. His special interests include data acquisition and image processing for multimedia, scientific, and business applications. He can be reached on AppleLink as MacSciTech or on the Internet or BIX at
editors@bix.com
.
Desktop Data Conferencing
