Periodogram and Pwelch of acceleration data

Hi all,

I am doing a frequency analysis of real signals. I have acceleration data from an impact hammer for six channels. I have calculated psd and Pwelch for each channel. The acceleration data was first filtered with zero phase shift Lowpass IIR Butterworth filter and then Psda and Pwelch are calculated. I have some questions regards my plots:

1.The plotted PSD and Pwelch are very noisy. Is there a way how to obtain more smooth curve for both PSD and Pwelch?

2. Also I have dominating peaks around 0.6Hz. Any idea how these can be removed?

I should mention that acceleration data shown here is cut from an interval that is much longer. Attached you can see the plotted psd and Pwelch and accelration daata used. Also you can see the code I have used.

LP IIR Butterworth:

Nf = 1;
Fpass = 70;
Apass = 1;
Astop = 30;
Fstop = 200;
% Design the filter using designfilt
LP_IIR_B = designfilt('lowpassiir','PassbandFrequency',Fpass,...
  'StopbandFrequency',Fstop,'PassbandRipple',Apass,'StopbandAttenuation',...
  Astop,'SampleRate',Acc_fs,'DesignMethod','butter');
% Filter the acceleration data by using filtfilt function in order to
% avoide phase shift of data
e = size(WB_Acc_S2_A2_R2);
WB_Acc_S2_A2_R2_Filt = zeros(e);
for i = 1:e(2);
WB_Acc_S2_A2_R2_Filt(:,i) = filtfilt(LP_IIR_B,WB_Acc_S2_A2_R2(:,i));
end

Periodogram:

%Number of samples
N = length(WB_Acc_S2_A2_R2_Filt); 
%Sampling frequency in Hertz
fs_A2 = Acc_fs; 
%Number of FFT points
NFFT = 2^nextpow2(N);
[Periodogram_S2_A2_R2,freq_A2_R2_Perio] = periodogram(WB_Acc_S2_A2_R2_Filt,[],...
  NFFT,fs_A2);

Pwelch:

N = length(WB_Acc_S2_A2_R2_Filt); 
%Sampling frequency in Hertz
fs_A2 = Acc_fs; 
%Number of FFT points
NFFT = N;
%Overlapping segments of samples
Noverlap = NFFT/3;
%Window length and type of window
win = NFFT;
wlen = rectwin(win);
%Calculate the psd using pwelch function
[PWelch_S2_A2_R2,freq_A2_R2_PWelch] = pwelch(WB_Acc_S2_A2_R2_Filt,...
  wlen,Noverlap,NFFT,fs_A2);

PS This is my first time calculating periodogram and Pwelch.

Many thanks,

Emina

Answer 1

[Brandon Eidson]

Hey Emina,

I understand that you are concerned about the noise and 0.6-Hz component in your resulting PSD and Periodogram plots. I checked out your attached figures.

It is my opinion that these are not abnormal features of frequency domain plots with even somewhat real-world data. The noise in your figures is extremely low; most of your y-axis values are below -200 dB/Hz!

Also, what you are seeing at 0.6 Hz is probably just a result of the DC component. There is no need to remove it.