“…We use a high-SNR approximation commonly made in the literature [25], [29]: R(δ δ δ,λ λ λ u ,λ λ λ b ) ≈ 1 2 ∑ i∈{u,b} ∑ N n=1 w i,n log 2 ( SNR i, n (δ δ δ, λ λ λ¯i ) ) (a) = − 1 2 log 2 ∏ i∈{u, b} ∏ N n = 1 (ISNR i, n (δ δ δ,λ λ λ¯i) w i,n . The term ISNR i,n in (a) denotes the inverse of SNR i,n , that is ISNR i,n = 1/SNR i,n .…”