Fig. 3.

SBL and l₁ sparseness metrics compared to the desired l₀ norm (dotted line). Each curve represents the sparseness metric for an arbitrary vector w with only K = 1, … , 5 non-zero coefficients at any position. All the non-zero weights are given the same magnitude r for different values of r on the x axis Ideally, we would like the sparseness metric to be inversely proportional to the the l₀ norm, which will be equal to the number of non-zero components (K) regardless of the value of the components themselves (i.e. r). Note that the SBL metric approximates better the l₀ norm, while l₁ norm deviates significantly from this ideal behavior.