Seperate high and low frequeny components from signal.

Hello

Please help me in separating low and high frequency components of signal(Time series data) recorded from a biological system. For data please refer to the column 2 of attached file.

The objective of my work is to select a random frequency threshold and then plot separate time series for low and high frequency components.

Thank you

[Star Strider]

Taking the fft of your data, essentially all the signal energy is below index 31.

What is the time base you are using (sampling frequency)? The first column of your data matrix is simply the index of samples, not a specific time, at least not that I can determine.

What do define as ‘high frequency’?

[Rajan]

I recorded this data at 10 kilohertz frequency.

and regarding 'high frequency':

The main objective of my work is to take power spectral density(psd) of the signal and plot it in log-log axes(log(power) vs log(frequency)) and see the slope of the plot using linear best fit, after that i need to select a threshold frequency and see whether or not slope is changing for low and high frequency components. if it is changing i need to seperate those components.

For data please refer to the column 1 of attached file and also PSD plot attached. you can see in figure, there are two components in signal (low and high frequency) which is having different slope. I need to seperate those components and make two different time series plot.

kindly help me..

thank you

[Star Strider]

Your figure will not display. The MathWorks website says it contains errors.

[Rajan]

Please check again I have updated the file.

[Star Strider]

When I did a FFT of your data (column 1), there was only noise above about 5 Hz.

Answer 1

[Image Analyst]

You can put this in a loop to extract from element 1 to the point and from the point to the end. Fit each segment to a line and record the slope. Plot the slopes and see where it starts to deviate from a constant. Something like (totally off the top of my head and untested)

for k = 1 : length(signal)
  leftSignal = signal(1:k);
  rightSignal = signal(k:end);
  % Now fit
  coeffs = polyfit(1:k, leftSignal, 1);
  leftSlopes(k) = coeffs(1);
  coeffs = polyfit(k:length(signal), rightSignal, 1);
  rightSlopes(k) = coeffs(1);
end
plot(leftSlopes, 'b-', 'LineWidth', 2);
plot(rightSlopes, 'r-', 'LineWidth', 2);
grid on;
legend('Left Slopes', 'Right Slopes');

[Rajan]

Thank you very much for your reply.

I am analyzing the slope after converting time series data to it's frequency domain by power spectral density analysis. The plot is in log(power) vs log(frequency). as you can see from attached figure the change in slope is in frequency domain, not in time domain.

So my question is how can I separate low frequency and high frequency components of my signal(time series data) which corresponds to different slope(frequency domain).

Thanks

[Image Analyst]

The signal I was talking about was the log of the frequency domain signal, NOT the time domain signal. So it should still work once you use the proper signal.

[Rajan]

Thanks for your reply.

Is there any method by which I can directly filter the original time series data? Because my final objective is to separate high and low frequency components from original data. And plot two different time series Where each corresponds to different slops.