Analog Coding Frame-work

TL;DR

This paper extends analog coding using frames, especially ETFs, to improve erasure resilience, and establishes a theoretical link between deterministic frames and random matrix spectra, with practical implications for coding and signal processing.

Contribution

It introduces a novel relation between deterministic frames and random matrix theory, and extends the Welch bound to erasure settings, demonstrating ETFs' advantages.

Findings

MANOVA ensemble describes spectra of random subframes.

Asymptotic moments of ETFs match MANOVA predictions.

Extended Welch bound shows ETFs' optimality in erasure scenarios.

Abstract

Analog coding is a low-complexity method to combat erasures, based on linear redundancy in the signal space domain. Previous work examined "band-limited discrete Fourier transform (DFT)" codes for Gaussian channels with erasures or impulses. We extend this concept to source coding with "erasure side-information" at the encoder and show that the performance of band-limited DFT can be significantly improved using irregular spectrum, and more generally, using equiangular tight frames (ETF). Frames are overcomplete bases and are widely used in mathematics, computer science, engineering, and statistics since they provide a stable and robust decomposition. Design of frames with favorable properties of random subframes is motivated in variety of applications, including code-devision multiple access (CDMA), compressed sensing and analog coding. We present a novel relation between deterministic frames and random matrix theory. We show empirically that the MANOVA ensemble offers a universal description of the spectra of randomly selected subframes with constant aspect ratios, taken from deterministic near-ETFs. Moreover, we derive an analytic framework and bring a formal validation for some of the empirical results, specifically that the asymptotic form for the moments of high orders of subsets of ETF agree with that of MANOVA. Finally, when exploring over-complete bases, the Welch bound is a lower bound on the root mean square cross correlation between vectors. We extend the Welch bound to an erasure setting, in which a reduced frame, composed of a random subset of Bernoulli selected vectors, is of interest. The lower bound involves moment of the reduced frame, and it is tight for ETFs and asymptotically coincides with the MANOVA moments. This result offers a novel perspective on the superiority of ETFs over other frames.