![[INTEL NAVIGATION HEADER]](../../../../DESIGN/PIX/HEADER.GIF){kind=small}
Written by
Dan Cox - IAL Media and Interconnect Technology Lab
dan_cox@ccm.jf.intel.com
&
Winnie Ng - DPG IHV Relations
Intel Corporation
No license, express or implied, by estoppel or otherwise, to any intellectual property rights is granted herein, and Intel disclaims all liability, including liability for infringement of any proprietary rights, relating to implementation of information in this document. Intel does not warrant or represent that such implementation(s) will not infringe such rights. Intel further disclaims any express or implied warranty relating to the sale and/or use of Intel products, including liability or warranties relating to fitness for a particular purpose or merchantability. * Other brands and names are the property of their respective owners.
This paper is addressed to IHVs and OEMs who have detailed working knowledge of the current PC audio architecture. It is also recommended that the reader be familiar with the Audio Codec '97 Component Specification available on the Intel Web server at http://www.intel.com/pc-supp/platform/ac97/.
This paper primarily describes AC '97's support for audio functionality in relation to "Multimedia" and "Communicating" PC functionality. Implementation specific details relating to modem data pump and phone line Codec design are left to the development community, and will not be covered in this paper.
The AC '97 architecture was designed to enable vendors who choose to source both the AC '97 controller and AC '97 Codec to deliver a highly integrated audio/telephony solution. However, at this point in time, the specification of interoperability for integrated audio/telephony solutions is beyond the scope of the AC '97 definition. Vendors who wish to achieve greater levels of interoperability amongst themselves may choose to work together. The industry may wish to move toward standardization of integrated audio/telephony interoperability. Feedback to this effect can be directed to Intel at audio97@intel.com.
Most voice modems employ two Codecs: one for an attached phone (DLP) and one for the Telco line. AC '97's mixer capabilities make it possible to eliminate the need for a separate phone Codec. And AC '97's optional modem DAC and ADC are designed to function as the line Codec. The optional DAC and ADC provide independent channels into the AC '97 controller, and require additional hardware support in the AC '97 controller (as well as additional software driver support).
Three telephony scenarios using AC '97 are discussed below.
For existing add-in voice modem designs with phone and line Codecs, AC '97's optional modem line DAC and ADC are unnecessary. In order to utilize system audio for headset or speakerphone, the following connections are needed:
An alternative architecture for add-in modems doesn't require any analog connections between the modem and AC '97. The modem still employs phone and line Codecs, but for headset or speakerphone the modem driver exchanges digital audio streams with the system audio driver via low latency ring 0 connections:
The entire telephony subsystem can be integrated onto the motherboard using the AC '97 design:
In this scenario AC '97 manages the DLP, mic and speakers (or headset), as well as modem line DAC and ADC. An integrated audio/telephony driver or low latency ring 0 connections are required to support full-duplex connections between system audio and the modem line Codec, which are logically independent drivers. Echo cancellation can be performed in the AC '97 digital controller hardware or in software on top of the system audio driver input.
There are two primary peripheral models for delivering speakerphone capabilities. The selection of the model has a direct impact on the cost, capabilities and limitations of the speakerphone implementation. At this point in time, a dedicated speakerphone model is used by room conferencing equipment, and a system audio speakerphone model is used for the PC.
In this model the speakerphone is implemented as a dedicated mono speaker/mic peripheral. In order to apply this model to the PC, the speakerphone unit might attach to the telephony card or be connected via USB.
PROs
CONs
In this model the speakerphone is implemented with the mic and speakers attached to the PC's system audio Codec.
PROs
Acoustic Echo Cancellation (AEC) filtering is a requirement for speakerphone functionality in full-duplex (simultaneous input and output) PC telephony and conferencing environments, and could become an exciting new capability for "hands free" DSVD games. AEC involves removing reflections of the output signal (which plays through the speakers) from the incoming signal (captured at the microphone). An echo cancellation filter requires digital representations of both the input and output data streams.
There are several forms of Echo Cancellation (EC) in the PC audio/telephony environment:
"Speakerphone AEC" and "stereo AEC" are discussed in the following sections.
The typical hardware or software echo canceller for "speakerphone"
telephony removes audio coupling caused by the far end voice playing
through the near end speakers and returning via the near end mic
(heard as echoes on far end). This requires 1 delay line
and 1 filter to remove the far end mono out (present in
both L out and R out) component from near end mic
in.
"Speakerphone AEC" is primarily designed as a "voice
only" echo canceller, and prevents talkers on the far end
from hearing echoes of their own voice caused by acoustic coupling
on the near end. This type of canceller employs an adaptive
filter and, being modeled on the dynamics of a two way conversation,
typically performs voice activity detection and
attempts to update the filter coefficients under the right conditions.
If continuous music or true stereo audio sources play, this can
interfere with adaptation.
The AC '97 architecture implements hardware support for the "speakerphone
AEC" function.
Depending on the actual filter implementation, there may or may not be benefit to adding L+R (or performing stereo to mono conversion) to get a better output reference signal in order to extend the cancellation capability to other than mono source material with a 1 delay line filter:
NOTE: For robustness, the filter coefficient adaptation model for this type of echo canceller should be less dependent on voice activity detection.
Robust DSVD games with speakerphone requires a new form of echo
cancellation. In a DSVD game scenario which employs speakerphone,
it would be highly desirable to additionally remove the near end
game's audio material (music, sound effects, explosions, etc...)
playing to speakers from what is transmitted via mic to the far
end (heard as interfering material on far end). This requires
2 delay lines and 2 filters to independently remove L out
from mic in and R out from mic in. Speech
recognition (for command and control) during the presence of stereo
output also benefits from the "stereo AEC" filtering.
The result is a mic input signal free from interfering stereo
output material.
NOTE: "Stereo AEC" differs from "speakerphone
AEC" due to the presence of continuous music or true stereo
audio sources, and requires a more sophisticated filter coefficient
adaptation model.
The AC '97 architecture implements hardware support for the "stereo AEC" function.
The current generation of host-based echo cancellers operate on
a data format known as "time correlated i/o".
This format is only meaningful when the Codec is operating in
full-duplex mode. The time correlated i/o format is a 2-channel
format which resembles the traditional interleaved stereo format.
Each sample, instead of containing left and right inputs, contains
an input (captured from mic) in the left channel and the current
output (destined for speakers) in the right channel.
If stereo PCM were playing out, the audio driver could add the left and right output samples (perform stereo to mono conversion) and interleave the sum with the incoming mono PCM samples. In this case, 2-channel time correlated data could be made available to the input stream for an enhanced mono "speakerphone AEC" function.
Today's multimedia Codecs generate or mix additional audio sources
(digital and analog) into the output signal (such as hardware
MIDI synthesis and analog Red Book CD audio). There is no way
to capture this output data at the software driver level. But
there are simple additions that can be made to the Codec mixer
and input MUX to support the capture of 2- or 3-channel input
data that is perfectly time correlated and at the desired input
sample rate.
For "speakerphone AEC" the L and R output channels can
be mixed together (in the analog mixer stage) to generate a mono
echo cancellation reference signal. Recording 2-channel data
from mic returns mic in in the left channel and the
Codec's internal generated mono out mix for the right
channel. This technique was recommended in Intel's Audio Hardware
Interface '96 Design Guide (a.k.a. Codec '96), and has been implemented
in several of the currently available 1-chip ISA Multimedia Codecs.
The baseline AC '97 mixer implements support for this capability.
For "stereo AEC" a 3-channel format is needed. The AC '97 analog mixer defines an optional 3rd ADC input which is dedicated to the mic, and can be used to return mic in along with the stereo out.
Current host based full duplex (FDX) speakerphone implementations,
such as Intel's controllerless DSVD modem and ProShareTM
Video Conferencing System, use an echo cancellation filter which
implements 1 delay line and 1 filter to remove the far end voice
from what is returned via the near end mic. Since hardware support
for software echo cancellation has not yet been widely implemented,
Intel's "speakerphone AEC" implementation uses a native
audio driver which supports software correlation. This echo cancellation
filter spans the output stream which gives it control over the
near end output volume (which is primarily needed in half-duplex
(HDX) mode before convergence or if the canceller is told to save
host MIPS by only doing HDX). Output volume control could also
have been implemented via Microsoft's Mixer API.
This AEC filter supports being informed of volume control changes
so as to track volume levels and minimize the need to reconverge
when levels are adjusted by the user (however, users who adjust
any speakerphone's volume control will typically notice
some artifact).
Currently, the limitations of current software audio drivers' ability to generate the needed echo cancellation reference signals is inhibiting the development of more advanced echo cancellers. Developing the next generation of speakerphone and stereo echo cancellers may depend on support from the audio Codec hardware.
V.34 is the ITU-T1 data modulation standard for modems operating
at speeds up to 28.8 Kbps. A revision of V.34 which allows two
new modem speed options, 31.2 Kbps and 33.6 Kbps, is expected
to be ratified in October 1996. V.34 is the modem datapump standard
required in DSVD 1.2 and V.70 modems, and H.324 terminals.
V.34 modems support both full duplex and half-duplex modes of operation and channel separation by echo cancellation techniques. Half-duplex mode will be used by high-speed FAX devices. V.34 modems support synchronous data signaling rates ranging from 2.4 Kbps to 28.8 Kbps (soon 33.6 Kbps) in multiples of 2.4 Kbps. The data signaling rate is determined by the modems during modem startup by measuring the transmission characteristics of the connection, and operating at the highest possible rate that will be supported by that particular connection.
V.70 is the ITU-T standard for DSVD modems. The standard supports
the simultaneous transmission of data and digitally encoded voice
signals over the POTS2 network. V.70 supports point-to-point connections
for DSVD operations. V.70 also supports multipoint connections
through a Multipoint Control Unit (MCU).
A V.70 system has the following characteristics:
V.80 is the ITU-T standard for the Synchronous Access Mode (SAM)
DTE3-to-modem protocol. V.80 is a new standard protocol for the
PC COM port. This standard is the key enabler of host based implementations
of communication protocols on PC's. Host based communication
protocols have to communicate with modem controllers and datapump
through the asynchronous PC COM port. V.80 enables the bi-directional
and simultaneous exchanges of commands and data between the PC
host and a synchronous modem datapump. V.80 converts the synchronous
bitstreams to asynchronous octets in order to pass the data through
the PC COM port. This process is called Synchronous Access Mode
(SAM).
V.80 allows host based software to implement new communications protocols. For example, a H.324 based video conferencing application is an example of applications that can utilize the V.80 standard. In most implementations, V.80 will not require a change to the modem hardware but rather an update to the modem controller firmware.
H.324 is the ITU-T standard for an analog multimedia communication
terminal which utilizes V.34 modems operating over the POTS network.
H.324 terminals may carry real-time voice, data, and video, or
any combination thereof, simultaneously over a high speed modem
connection.
H.324 terminals can be integrated into personal computers or implemented
in stand-alone devices such as videophones. Support for each
media type (voice, data, audio) is optional. H.324 supports simultaneous
use of multiple channels of media types.
Modems used in H.324 terminals must operate in full-duplex, in either synchronous mode or V.80 synchronous access mode, and conform to ITU-T V.34 and ITU-T V.8 standards; support of V.8bis is also required.
__________________________________
1 International Telecommunication Union - Telecommunication Standardization Sector
2POTS - "Plain Old Telephone System"
3DTE - Data terminal equipment. It can be the host PC or modem hardware depending on implementation.
In order to provide their customers with cost-effective communication solution such as video conferencing, many vendors have or are working on modem products with host based modem controllers (hardware data pump). AC '97 can support the audio Codec hardware requirements to provide functionality of plain data, voice and data, and video, voice and data modems.
With the introduction of more powerful processors and Intel's MMXTM technology, some vendors are working on host based modem solution (soft modem). AC '97 can support the audio Codec requirements of a soft modem implementation by providing the i/o functionality required in modem communication environment.
* Legal Stuff © 1997 Intel Corporation
