EE Seminar: Source-Channel Separation is Not Too Bad

~~Speaker: Uri Mendelovic,
M.Sc. student under the supervision of Prof. Meir Feder

Wednesday, March 2, 2016 at 15:00
Room 011, Kitot Bldg., Faculty of Engineering

Source-Channel Separation is Not Too Bad
Abstract

Shannon's source-channel separation theorem, originally proved for point-to-point communication problems, was refuted for many network problems. In this work we revisit separation and show some practical considerations in favor of separation in network problems. Two major problems are considered: broadcasting a source with distortion to two or more receivers and collaborative transmission over asynchronous multiple-access-channel (MAC). For both problems, we study the general case and then provide a tighter analysis for the AWGN case. Upper bounds on the gap between separation-based coding and the optimal performance are presented.
For degraded broadcast channel, separation-based time-sharing achieves a non-zero factor of the joint source-channel optimal rate, regardless of number of receivers. For the AWGN broadcast channel a better bound is introduced, implying that all achievable joint source-channel schemes have a rate within one bit of the separation-based achievable rate region for two receivers, or within log2T bits for T receivers.
For the MAC channel, separation holds when transmitting independent sources. When sources are correlated, or in the extreme case of fully cooperative transmitters, a better performance may be achieved by joint source-channel coding. This gap can be arbitrarily large, but joint coding requires perfect transmitter synchronization. We define totally asynchronous transmitters and prove that in this case collaboration has no performance gain.