Algorithm 3. GDBRA

○ Step 1: For each k = 0, 1, …, N − 1 calculate generalized deviation GD(k) as: $$\begin{array}{l}GD(k)=\hfill \\ =\underset{n_m\in N_{avail}}{mean}\left\{{\left|x(n_m)e^{-j2\mathit{\pi }k{n}_m/N}-\underset{n_m\in N_{avail}}{mean}\left\{x(n_1)e^{-j2\mathit{\pi }\frac{k{n}_1}{N}},\dots ,x(n_M)e^{-j2\mathit{\pi }\frac{k{n}_M}{N}}\right\}\right|}^L\right\},\hfill \end{array}$$ where $\underset{n_m\in N_{avail}}{mean}\{f(n_M)\}$ is used to find mean value of vector with elements $f(n_M)$, for $n_M\in Navail$. For norm ℓ2 use L = 2, while for norm ℓ1 use L = 1. ○Step 2: Determine the signal support $k=\arg \{GD(k)<T\}$ where T can be calculated with respect to $median\{GD(k)\},$ or $p\cdot median\{GD(k)\}$ (where p is constant close to 1) for example, but also with respect to mean or minimal value. The vector of positions k should contain all signal frequencies $k_i\in k$ for any I = 1, …, K. ○Step 3: Set $\tilde{X}(k)=0$ for frequencies $k\notin k;$; ○ Step 4: The estimates of the DFT values can be calculated by solving the set of equations, at the localized frequencies from the vector k, where k contains K signal frequencies k = k1, k2, … kk. A system of equations is set as follows: $${\displaystyle {\sum }_{i=1}^{K}X(k_i)e^{j2\mathit{\pi }{k}_i{n}_m}}=x(n_m)$$ ○ Step 5: The CS matrix ACS is formed as a partial DFT matrix: columns correspond to the positions of available samples, rows correspond to the selected frequencies. The system is solved in the least square sense: $$X={\left(A_{CS}^T{A}_{CS}\right)}^{-1}{A}_{CS}^{T}y.$$