PARALLEL CHANNELS WITHOUT CROSSTALK
Abstract
In this report, a study is made of information theoretic channels which are decomposable into a number of parallel subchannels which will, in general, be dependent. For this situation, two models are constructed in which each subchannel input affects only the corresponding subchannel output (no crosstalk). In the first model (MC channel), the lack of crosstalk is ensured by constraints on the channel conditional probability distribution. The second model (MS channel) is a channel with an underlying state structure with states independent of the input. Both models are memoryless. All MS channels are MC, but the reverse does not hold. The effect of subchannel dependencies on capacity and random coding exponent (RCE) is investigated. It is proved that these dependencies cannot decrease the capacity of our channels. However, subchannel dependencies may either increase or decrease the RCE. It is also proved that the capacity of the channel is not less than the sum of the capacities of the individual subchannels. When the state model is used, the above two quantities are equal if the receiver has knowledge of the channel state. A definition of partial state knowledge is given. It is proved that, when the receiver has partial state knowledge, the resulting capacity and RCE are not decreased. For complete state knowledge at the receiver, the capacity and RCE are not less than those obtained for partial state knowledge. Various coding alternatives are discussed, and formulas are given for computing or bounding performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 09, 1968
- Accession Number
- AD0673479
Entities
People
- Joel Max
Organizations
- Massachusetts Institute of Technology