fix(dsp): upconvert and downconvert matrix return matrix (#38)

src/dsp.jl

@@ -123,7 +123,7 @@ end
 function upconvert(s::AbstractMatrix, sps, fc, pulseshape=rrcosfir(0.25, sps); fs=framerate(s))
  mapreduce(hcat, eachcol(s)) do x
  upconvert(x, sps, fc, pulseshape; fs=fs)
- end
+ end[:,:]
 end
 
 """
@@ -143,7 +143,7 @@ end
 function downconvert(s::AbstractMatrix, sps, fc, pulseshape=rrcosfir(0.25, sps); fs=framerate(s))
  mapreduce(hcat, eachcol(s)) do x
  downconvert(x, sps, fc, pulseshape; fs=fs)
- end
+ end[:,:]
 end
 
 """
@@ -499,19 +499,16 @@ function _istft(iX::AbstractMatrix{T}, nfft::Int, noverlap::Int, window::Union{F
  # H. Zhivomirov, TEM Journal, Vol. 8, No. 1, pp. 56-64, 2019.
  (window === nothing) && (window = rect)
  win, norm2 = Periodograms.compute_window(window, nfft)
-
  nstep = nfft - noverlap
  nseg = size(iX, 2)
  outputlength = nfft + (nseg-1) * nstep
-
  iX .*= win
  x = zeros(T, outputlength)
  normw = zeros(eltype(win), outputlength)
  for i = 1:nseg
  @views x[1+(i-1)*nstep:nfft+(i-1)*nstep] .= x[1+(i-1)*nstep:nfft+(i-1)*nstep] .+ iX[:,i]
  @views normw[1+(i-1)*nstep:nfft+(i-1)*nstep] .= normw[1+(i-1)*nstep:nfft+(i-1)*nstep] .+ win .^ 2
  end
-
  trimlength = nfft % 2 == 0 ? outputlength - nfft : outputlength - nfft + 1
  (sum(@view(normw[1+nfft÷2:end-nfft÷2]) .> 1e-10) != trimlength) && (
  @warn "NOLA condition failed, STFT may not be invertible")

test/runtests.jl

@@ -432,6 +432,16 @@ end
  @test framerate(z) == 100
  @test amp2db(rms(x-z)/rms(z)) < -40.0
 
+ x = x[:,1:1]
+ y = upconvert(x, 10, 100)
+ @test nchannels(y) == 1
+ @test framerate(y) == 1000
+ @test y isa AbstractMatrix
+ z = downconvert(y, 10, 100)[12:end-11,:]
+ @test nchannels(z) == 1
+ @test framerate(z) == 100
+ @test z isa AbstractMatrix
+
  x = rcosfir(0.25, 10)
  @test length(x) == 221
  @test x[1:10:end] ≈ vcat(zeros(11), 0.326598866403003, zeros(11))