How Rate-Limited Feedback Increases Capacity for Memoryless Broadcast Channels
Event details
| Date | 03.02.2014 |
| Hour | 14:15 › 16:00 |
| Speaker |
Prof. Michèle Wigger, Telecom Paris Bio: Michèle Wigger received the M.Sc. degree and the Ph.D. degree both in electrical engineering from ETH Zurich in 2003 and 2008, respectively. After her PhD, she spent 7 months as a postdoctoral researcher at the Information Theory and Applications (ITA) center, University of California, San Diego. Since December 2009, she has been an assistant professor at Telecom Paris-Tech (former ENST), Paris, France. |
| Location | |
| Category | Conferences - Seminars |
We present new coding schemes for the two-receiver discrete memoryless broadcast channel (BC) with rate-limited feedback from one or two receivers. With an appropriate modification, our schemes also apply to the setup with noisy feedback when the receivers can code over the feedback links. Our first scheme strictly improves over the nofeedback capacity region for the class of strictly essentially less-noisy BCs, even when there is only feedback from the weaker receiver and no matter how small (but positive) the feedback rate. Examples of essentially strictly less-noisy BCs are the binary symmetric BC or the binary erasure BC with unequal cross-over probabilities or unequal erasure probabilities to the two receivers. Our scheme also improves over the nofeedback capacity region of the binary symmetric/binary erasure BC and also for parameter ranges where the BC is not essentially less-noisy but more capable. So far, feedback was known to increase capacity only for a few very specific memoryless BCs with feedback.
The second scheme that we present has the following property. When the feedback-rates are sufficiently large, then it can recover all previously known capacity and degrees of freedom results for memoryless BCs with feedback, i.e., the results by Dueck, Wang, Georgiadis and Tassiulas, Shayevitz and Wigger, and Maddah-Ali and Tse. In particular, as the feedback-rates tend to infinity our scheme converges to a special case of the Shayevitz-Wigger scheme which is known to include the other schemes as special cases.
We further show a duality relationship between the set of rates achieved with linear-feedback schemes over multi-antenna memoryless Gaussian MACs and BCs. In this part we assume that the feedback is perfect, i.e., of infinite rate. When either transmitters or receiver are single-antenna, the capacity region of the Gaussian MAC with feedback is known and is achieved by a linear-feedback scheme. For these special cases our result thus allows us to determine the set of rates that can be achieved over the Gaussian BC with linear-feedback schemes, and to identify the optimal linear-feedback schemes.
Joint work with Selma Belhadj Amor, Yossef Steinberg, and Youlong Wu.
The second scheme that we present has the following property. When the feedback-rates are sufficiently large, then it can recover all previously known capacity and degrees of freedom results for memoryless BCs with feedback, i.e., the results by Dueck, Wang, Georgiadis and Tassiulas, Shayevitz and Wigger, and Maddah-Ali and Tse. In particular, as the feedback-rates tend to infinity our scheme converges to a special case of the Shayevitz-Wigger scheme which is known to include the other schemes as special cases.
We further show a duality relationship between the set of rates achieved with linear-feedback schemes over multi-antenna memoryless Gaussian MACs and BCs. In this part we assume that the feedback is perfect, i.e., of infinite rate. When either transmitters or receiver are single-antenna, the capacity region of the Gaussian MAC with feedback is known and is achieved by a linear-feedback scheme. For these special cases our result thus allows us to determine the set of rates that can be achieved over the Gaussian BC with linear-feedback schemes, and to identify the optimal linear-feedback schemes.
Joint work with Selma Belhadj Amor, Yossef Steinberg, and Youlong Wu.
Practical information
- Informed public
- Free
Organizer
- IPG Seminars
Contact
- Host: LINX ([email protected])