FFT single-sided amplitude spectrum plot spike at 1Hz of accelerometer data

i m trying to plot the alpitude spectrum of Z-axis accelerometer with a sample frequency of ~50Hz and signal length 18429.

Fs=46.9350530441355
% t is the time vector in millisecond from 0 to 392648.965000002
L=length(y) % y is accelerometer data 18429 records
subplot(2,1,1)
plot(t,y)
title('Signal')
xlabel('time (milliseconds)')
NFFT = 2^nextpow2(L); % Next power of 2 from length of y
Y = fft(y,NFFT)/L;
f = Fs/2*linspace(0,1,NFFT/2+1);
subplot(2,1,2
% Plot single-sided amplitude spectrum.
plot(2*abs(Y(1:NFFT/2+1))) 
title('Single-Sided Amplitude Spectrum of y(t)')
xlabel('Frequency (Hz)')
ylabel('|Y(f)|')

What i get on second plot it starts with x(1), y(18.46) and than fading. the data are from accelerometer atached to car windsheald.

[Wayne King]

What is your question?

[DuckDuck]

My question is that i don't understand what the plots are telling me!!

Why i have this spike of magnitude at 1HZ?

[Dr. Seis]

You aren't plotting the amplitude spectrum against your frequencies "f", so how do you know the spike is at 1Hz? Your frequency increment (df) is Fs/NFFT, which is approximately 46.935/32768 = 0.0014. So your first non-zero frequency should be 0.0014 Hz, not 1Hz.

Regardless... as this is acceleration data, you highest amplitudes should be at higher frequencies. Is your data superimposed on a low-frequency sine wave? Or does it have a linear trend associated with it?

[DuckDuck]

my data are raw data from phone accelerometer measuring road surface . i then performed a hamming window over the framed data, 250 samples per frame overlapping each other.

Ok about frequency scale, but how come it spikes at lower frequencies?

[Dr. Seis]

See my previous questions... are your high-frequency data superimposed on a very low-frequency sinusoidal wave? Or are your high-frequency data superimposed on a linear trend?

Can you load a picture of what your time data look like?

[DuckDuck]

this is the plot of my data http://img18.imageshack.us/img18/3065/plotp.jpg

[Dr. Seis]

Ah... yeah, you should remove the mean before you run it through the FFT. Aside from that, there are some low-frequency bits in there, but none that should compare to the mean. Your first amplitude in Y is at zero-frequency, which is the mean of the data.

[DuckDuck]

thank you for your help. now i get this picture for the first window of fft data: http://img137.imageshack.us/img137/8308/firstwindowfftmagnitude.jpg

how can i make in matlab a spectrogram like this one http://img37.imageshack.us/img37/8367/spectrogram.jpg

now that i have the fft transformed data?

[Dr. Seis]

I would create a new question... others who are more familiar with running the Matlab function "spectrogram" and overlaying plots will be able to help you. You do not need to transfer your time data into the frequency domain before you use "spectrogram" - you will input your time data directly.

Answer 1

[Dr. Seis]

As a general rule of thumb, I will remove any trend and/or mean from the data before I run it through FFT.