CODING THEOREMS FOR FILTERED-WAVEFORM CHANNELS,
Abstract
Filtered-waveform channels are continuous-time channels with waveform inputs satisfying a power constraint; the inputs are passed through a linear filter and corrupted by additive stationary Gaussian noise. Both a white-noise channel and a colored-noise channel with a matched filter are investigated. There exist equivalent channels (both of exactly the same form) which transmit coefficients of orthogonal expansions defined by the channel and are specified by its normal values. The direct half of the coding theorem is proved by Feinstein's fundamental lemma, using the asymptotic distributions of normal values and power distributions upon the input random variables. Weak and strong converses are proved. The capacity is also shown to be the supremum of asymptotic time-average mutual information over a class of stationary Gaussian inputs with spectral densities Orthogonal expansions in terms of normal functions in Hilbert spaces isometric to random signal and noise permit expression of mutual information in terms of normal values whose asymptotic distribution is found. The coding theorem is proved by random coding using the measure of the input process. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1970
- Accession Number
- AD0708827
Entities
People
- John Kingholm Moore
Organizations
- University of Illinois Urbana–Champaign