An indirect Pulse-based Time-of-Flight camera can be modelled as a linear sensing system in which the target's depth is recovered from few measurements through a sensing matrix formed by a set of demodulation functions. Each demodulation function is the result of the convolution of a (0,1)-binary code and a cross-correlation function which models the entire modulation-demodulation process. In this paper, we present a practical scheme for the construction of the sensing matrix which relies on the optimization of the coherence, and is based on low-density codes. We demonstrate that our methodology eliminates the intrinsic variability of random and pseudo-random approaches, and allows for the recovery of the target's depth in a grid much finer than the number of distinct elements in the binary codes.