Welch PSD for tremor has very high frequency

I am trying to compute a PSD using the Welch Method on a tremor signal from a mouse. usually these tremors occur between 1Hz-15Hz but after doing various PSDs and FFTs nothing will come back with a dB/Hz within the range (usualy 50+) which is not only unusual but almost impossible. I have never done a PSD or FFT before so I taught myself from scratch. Can anyone help me figure out what I am doing wrong? here is my Matlab code:

% Load and extract signal and time from CSV

data = readmatrix('M2_C2823722_MatLab.csv');

time = data(:,1);

signal = detrend(data(:,2) - mean(data(:,2))); % Demean and detrend in one line

% Sampling frequency

fs = 1 / mean(diff(time));

% Apply Hamming window directly

signal = signal .* hamming(length(signal)); % No transpose needed

% Compute PSD with Welch method

window = 3000; % 3-second segments

noverlap = 1500; % 1.5-second overlap

[pxx, f] = pwelch(signal, window, noverlap, [], fs);

% Plot PSD

figure;

plot(f, 10*log10(pxx));

xlabel('Frequency (Hz)');

ylabel('Power/Frequency (dB/Hz)');

title('PSD Estimate: 3-Second Segments, 1.5s Overlap');

grid on;

% Save results

writetable(table(f, pxx), 'psd_resultsM2_3722.csv');

saveas(gcf, 'm2_3722.png');

[David Goodmanson]

Hi Jason,

the call to pwelch appears to be all right. Are the times in the data file equally spaced, or very close to it?

[Nathan]

Hi Jason,

If you use an integer for the 'window' input to pwelch, it internally uses a Hamming window on the chunks that it divides your signal into for averaging. Windowing it beforehand may cause issues; a global windowed signal chunked up into overlapping segments that are locally windowed may have weird effects.

In general, you should avoid windowing the entire signal before it gets chunked (assuming it's a vector); one purpose of windows is to manage the transitions between the overlaps when averaging the periodograms. If you want to specify the window that pwelch uses, you can do 'pwelch(signal, hamming(length(signal)), ...)', replacing hamming() with another window function, etc.

If you remove the line 'signal = signal .* hamming(length(signal));' and that doesn't fix it, try to replicate without any windowing or chunking in a very basic spectrum plot as a sanity check; if that still gives you odd frequencies, then we can look at other causes

spect_twoside = 10*log10(abs(fftshift(fft(signal)./fs)));
spect_oneside = spect_twoside(ceil(length(signal)/2):end);

f = freqspace(length(signal)).*fs/2;

plot(f, spect_oneside)

[Mathieu NOE]

@Jason

could you please provide the data file ? use the paperclip button

[Mathieu NOE]

@Jason

problem solved ?