difference in transfer function calculation in matlab and python

i have developed below code in matlab and trying implement samething in python

but getting different outputs from matlab and python.

can you help here to convert this code to python

Matlab code

[Y, freq] = tfestimate(in, out, hanning(NFFT), round(overlap*NFFT), [], sample_rate);

example output

freq:freq= [0;0.0101800266860033]

Python code

f, Pxy = csd(out_signal,in_signal, fs=sample_rate, window=windows.hann(NFFT, sym=True), average='median', nperseg=NFFT, noverlap=int(overlap * NFFT), scaling='spectrum')

example output

freq: [0; 0.0132058240]

[Mathieu NOE]

hello

seems to me you're comparing apples and oranges

csd gives the cross power spectral density , not a transfer function estimate

fyi : this is the code (of an older version) of tfestimate ; in the litterature this is called the H1 estimator and the Txy transfer function uses Pxx and Pxy :

for zero extra fee you get also the coherence as you have also Pyy :

%%%%%%%%%%%%%%%%%%%%%%%
function [Txy,Cxy,f] = mytfe_and_coh(x,y,nfft,Fs,window,noverlap)
% Transfer Function  and Coherence Estimate 
    
% compute PSD and CSD
window = window(:);
n = length(x);		% Number of data points
nwind = length(window); % length of window
if n < nwind    % zero-pad x , y if length is less than the window length
    x(nwind)=0;
    y(nwind)=0;  
    n=nwind;
end
x = x(:);		% Make sure x is a column vector
y = y(:);		% Make sure y is a column vector
k = fix((n-noverlap)/(nwind-noverlap));	% Number of windows
					% (k = fix(n/nwind) for noverlap=0)
index = 1:nwind;

Pxx = zeros(nfft,1); 
Pyy = zeros(nfft,1); 
Pxy = zeros(nfft,1); 

for i=1:k
    xw = window.*x(index);
    yw = window.*y(index);
    index = index + (nwind - noverlap);
    
    Xx = fft(xw,nfft);
    Yy = fft(yw,nfft);
    Xx2 = abs(Xx).^2;
    Yy2 = abs(Yy).^2;
    Xy2 = Yy.*conj(Xx);
    Pxx = Pxx + Xx2;
    Pyy = Pyy + Yy2;
    Pxy = Pxy + Xy2;
    
end

% Select first half
if ~any(any(imag([x y])~=0))   % if x and y are not complex
    if rem(nfft,2)    % nfft odd
        select = [1:(nfft+1)/2];
    else
        select = [1:nfft/2+1];   % include DC AND Nyquist
    end

    Pxx = Pxx(select);
    Pyy = Pyy(select);
    Pxy = Pxy(select);
    
else
    select = 1:nfft;
end
Txy = Pxy ./ Pxx;                   % transfer function estimate 
Cxy = (abs(Pxy).^2)./(Pxx.*Pyy);    % coherence function estimate 

f = (select - 1)'*Fs/nfft;

end

[siva prasad]

Thank you for the suggestion.

This is the code received from client and asked to convert it into python code.

I have used above python code along with welch for pxx. but when i compare results of

matlab code - freq

and python code - freq, i am getting some differences. so Please help me on converting to python.

[Mathieu NOE]

hello again

as I said before , the python code you mention is not the transfer estimate

see GitHub - giuliovv/tfest: Estimate transfer function from its frequency response

and signal analysis - Transfer Function Estimate - Signal Processing Stack Exchange