 |
 Analysis/Minimization of the Impact of Errors on Compressed Image & Video
Multimedia Multicasting
Concealment of the Time-Varying Impairments
|
An overall goal of this work is the generation of strategies to enable adaptive multimedia applications which gracefully and seemlessly adjust to the dynamic network and physical channel environments with minimal user involvement.
|
| A) |
Analysis/Minimization of the Impact of Errors on Compressed Image & Video
We are investigating the impact of transmission errors on interactive multimedia applications over wireless communications networks and are developing schemes for efficient error control. Even though retransmission (ARQ) protocols can be used to (eventually) achieve error free transmission of data, retransmissions increase the delay and the channel bandwidth utilization diminishes. User interactivity requirements impose specific delay constraints and thus a limit to the number of retransmissions. Forward Error Correction (FEC) can also be used to improve the channel characteristics, but this also introduces overhead and cannot by itself solve the problem completely.
Our starting point is the observation that "multimedia" is useful at various levels of quality. Therefore, unlike the case of traditional data communication, some types of data loss can be tolerated by users. On the other hand, transmission errors can be devastating for compressed media. This is due to the design assumptions of the most prevalent image and video compression schemes, such as JPEG and MPEG-1, which were targeted for error-free channels and non- interactive communications.
We are developing a new layered coding-based image/video transmission scheme which spreads the signal energy throughout all the layers of the coded image so that loss in any segment of the transmitted image does not seriously degrade the perceived signal quality. This approach effectively compensates for errors in various regions of an image and in situations where traditional error concealment schemes based on the interpolation of pixels in adjacent blocks fail.
Although this research is certainly coupled to that discussed in Section 3C, it is focused at problems arising at the higher layers and is reflective of not only channel impairments but also network impairments. Here, these issues shall be addressed in a fashion that is much closer to the applications.
A key question is the appropriate use of (re-)synchronization information and techniques and the investigation of the tradeoffs that arise. For example, the JPEG standard specifies an RST marker to limit error propagation, but its use has not been specified. Similarly, the selection of I, P, and B frame ratios in MPEG-1 generate interesting trade-offs, in particular if different types of frames are protected with different amounts of coding. Finally, protection for the re- synchronization information is critical and probably requires careful integration with the packetization schemes. This effort will continue through 4Q99.
|
[top of page]
|
| B) |
Multimedia Multicasting
Multicast refers to communication between a single sender and multiple receivers. Multicast is becoming an important service for communication networks, particularly when multimedia communication is involved. Because of the potential explosive resource demands and the opportunity for economies and performance improvements, it is important to design this service well.
Wireless channels with mobile stations exacerbate the problem. We are evaluating multicast architectures and developing new schemes that address these problems. In particular, we plan to investigate: (i) multicast in Mobile IP and the efficiency of the various alternative ways of supporting this service, and (ii) the use of broadcast capabilities of wireless channels to efficiently support multicast; note, however, that unlike traditional wireline LANs, channel errors and transmit power considerations limit the effectiveness of broadcasting and introduce various trade-offs.
In the case of interactive image and video transmission, hierarchical coding can facilitate multicasting by enabling destinations to adjust the quality of signal they each receive, independently of each other and without the source actually being aware of the adjustment. A similar function can be provided by intelligent network switches at critical points in a network where abrupt changes in available capacity occur (e.g., at the periphery of a fiber optic ATM network accessing a wireless, lower speed channel). We are developing an architecture for multicasting continuous media across heterogeneous networks and to heterogeneous end devices. This work will continue through 3Q99.
|
[top of page]
|
| C) |
Concealment of the Time-Varying Impairments
We will investigate new system software architectures that support adaptable communication channel abstractions. Given the expected proliferation of wireless links, both in comprising wireless networks and in being integrated with existing wired/wireless networks such as the Internet, providing communication abstractions for programmers that take into account the characteristics of wireless links is important. In particular, wireless links are distinguished by characteristics that are very different, and in some cases widely time- varying, compared to wired links: physical impairments (e.g., error rates change depending on location and environmental conditions), spectrum (e.g., relatively low compared to wired networks), location of nodes (e.g., mobile computers), and power requirements (e.g., depending on proximity of receiver and competition for spectrum).
One goal of system software (i.e., the operating system and network protocol layers between the hardware and the application) is to provide communications abstractions for interprocess communications on the same or on different machines. These communication abstractions are generally designed to be conceptually easy to understand. For example, the "pipe" abstraction has the characteristics of a stream of bytes (no visible message boundaries) that is reliable (no missing data). A real-time continuous media (audio or video) channel abstraction would support the delivery of messages (blocks of data representing samples from the analog source) by certain predetermined deadlines while maintaining an average throughput.
When the underlying physical communication medium is composed of one or more wireless links, the system software may be designed to automatically compensate for their impairments so that the desired communication abstraction is achieved. However, this automatic adjustment may introduce costs that the user may not be willing to accept. For example, to achieve reliability, the system software may retransmit packets. However, for an interactive multimedia application such as video-conferencing, which can tolerate errors such as a missing frame every so often but cannot tolerate high delays, it is more important to increase the probability that most packets will arrive by their expected deadlines than it is to achieve the successful arrival of all packets regardless of delay.
Consequently, we believe it is important that applications be given the ability to control the characteristics of the communication abstraction, which will be a function of their requirements and the expected performance characteristics of the underlying physical communication medium rather than simply having to accept a fixed set of characteristics. The need for adaptability is emphasized by the time-varying characteristics of wireless links. This ability to dynamically control the communication channel is difficult to do today because of the architecture and interface provided by most system software. Regarding architecture, most system software is not structured to support communication channels that will change over time (especially after a connection is established). Consequently, there is no provision at the interface to support dynamic modification of the channel. We propose to investigate new system software architectures that support dynamic control.
We are currently investigating how to best allow distributed client/server applications to control the power of a wireless connection. Consider a user with a handheld communication/computing device that communicates over a wireless line. Depending on whether this user is in an area where there are many other devices competing for the spectrum (e.g., on the floor of a convention hall), or in an area where there is no other wireless traffic (e.g., in the middle of the desert), the device may wish to adjust its power (in a cooperative way with other devices) to increase overall performance. The application knows best what its requirements are, based on its needed quality-of-service. To avoid burdening the application writer with having to understand how to control the specific underlying channel, all the application would do is provide its requirements to library software functions that turn these specifications into dynamic controls of the communication deviceÕs characteristics. This project, was started in September 1995 and its expected term is approximately three years.
|
|
Professors Joe Pasquale and George Polyzos conduct research in system support for multimedia communications in the Computer Systems Laboratory
|
[top of page]
|
|
|
|
 |
|
Affiliated Faculty
Research Thrust Areas