Through the Haze: a Non-Convex Approach to Blind Gain Calibration for Linear Random Sensing Models

TL;DR

This paper introduces a non-convex optimization approach using projected gradient descent for blind gain calibration in linear random sensing models, providing theoretical guarantees on convergence and sample complexity, even with noise.

Contribution

It develops a novel non-convex method for blind gain calibration with provable convergence and reduced sample complexity, applicable with or without priors.

Findings

Algorithm converges to the true solution under certain sample complexity conditions.

Sample complexity grows linearly with the number of unknowns, up to log factors.

Method performs well in noisy settings, with graceful degradation of accuracy.

Abstract

Computational sensing strategies often suffer from calibration errors in the physical implementation of their ideal sensing models. Such uncertainties are typically addressed by using multiple, accurately chosen training signals to recover the missing information on the sensing model, an approach that can be resource-consuming and cumbersome. Conversely, blind calibration does not employ any training signal, but corresponds to a bilinear inverse problem whose algorithmic solution is an open issue. We here address blind calibration as a non-convex problem for linear random sensing models, in which we aim to recover an unknown signal from its projections on sub-Gaussian random vectors, each subject to an unknown positive multiplicative factor (or gain). To solve this optimisation problem we resort to projected gradient descent starting from a suitable, carefully chosen initialisation point. An analysis of this algorithm allows us to show that it converges to the exact solution provided a sample complexity requirement is met, i.e., relating convergence to the amount of information collected during the sensing process. Interestingly, we show that this requirement grows linearly (up to log factors) in the number of unknowns of the problem. This sample complexity is found both in absence of prior information, as well as when subspace priors are available for both the signal and gains, allowing a further reduction of the number of observations required for our recovery guarantees to hold. Moreover, in the presence of noise we show how our descent algorithm yields a solution whose accuracy degrades gracefully with the amount of noise affecting the measurements. Finally, we present some numerical experiments in an imaging context, where our algorithm allows for a simple solution to blind calibration of the gains in a sensor array.