feat: different number of ref modes than tx modes

examples/sdm_channel_stats.jl

@@ -91,8 +91,8 @@ end
 """
 
 """
-function generate_reference_modes(midx, domain, λ, w0)
-    modes = zeros(ComplexF64, domain.Nx, domain.Ny, length(midx))
+function generate_reference_modes(midx, domain::Domain{T}, λ, w0) where {T}
+    modes = zeros(Complex{T}, domain.Nx, domain.Ny, length(midx))
     x = domain.x(domain.Nz)
     yt = transpose(domain.y(domain.Nz))
 
@@ -114,7 +114,7 @@ transfer_matrix!()
     The input arrays uin and uout are three dimensional arrays, where the different modes are concatenated along the third dimension.
 
 """
-function transfer_matrix!(H, uin::AbstractArray{T1,3}, uout::AbstractArray{T2,3}) where {T1,T2}
+function transfer_matrix!(H, uout::AbstractArray{T1,3}, uin::AbstractArray{T2,3}) where {T1,T2}
     for n ∈ axes(uin, 3)
         for m ∈ axes(uout, 3)
             @views H[m, n] = overlap_integral(uout[:, :, m], uin[:, :, n])
@@ -127,7 +127,7 @@ mdl_statistics()
 
 """
 
-function sdm_statistics(fn; basedir=pwd(), Nopt=10, statfun=mdl, target=mean, show_trace=false)
+function sdm_statistics(fn; ng=nothing, basedir=pwd(), Nopt=10, statfun=mdl, target=mean, show_trace=false, showplot=false)
 
     fn = joinpath(basedir, fn)
 
@@ -136,7 +136,11 @@ function sdm_statistics(fn; basedir=pwd(), Nopt=10, statfun=mdl, target=mean, sh
         read(h5["RXFields"]), Dict(attrs(h5))
     end
 
-    midx = values.(metadata["modes"])
+    midx = if isnothing(ng)
+        values.(metadata["modes"])
+    else
+        midx = sort!([(m, n) for m ∈ range(0, ng - 1) for n ∈ range(0, ng - m - 1)], by=sum)
+    end
 
     λ = metadata["lambda"]
 
@@ -148,19 +152,21 @@ function sdm_statistics(fn; basedir=pwd(), Nopt=10, statfun=mdl, target=mean, sh
     dz = metadata["dz"]
     w0 = metadata["w0"]
 
-    domain = Domain(dx[end], dy[end], dz, Nx, Ny, Nz)
+    T = eltype(dx)
+    domain = Domain{T}(dx[end], dy[end], dz, Nx, Ny, Nz)
 
     # TODO: we have to optimize w0, but over how many iterations? Convergence analysis? Too many iterations seem to make the optimization quite challenging and might result in bad convergence
     # reference_modes = optimize_reference_modes(view(urx, :, :, :, 1:Nopt), midx, domain, λ; target, show_trace)
     # lazy or more correct? Consider the analytical LG modes in distance domain.Dz with the given w0
     reference_modes = generate_reference_modes(midx, domain, λ, w0)
 
-    # TODO: idea of Sebastian: investigate rank of the transfer matrix
     stat = tf_stats(urx, reference_modes, statfun)
 
     @info statfun
     summarystats(stat) |> display
 
-    x, y = estimate_ccdf(stat)
-    plot.scatter(; x, y)
+    if showplot
+        x, y = estimate_ccdf(stat)
+        plot.scatter(; x, y)
+    end
 end
\ No newline at end of file