Averaging lots of columns at once and then combining them

I have 2 and a half minutes of EMG data that I need to conduct a short time fourier transform on and then average within the short 250 ms epochs. Then that data needs to be combined into 20 second intervals. I used the buffer funtion to separate the data into 250 ms intervals but I cannot figure out a way to take the average of each coloumn without manually inputing it. Is there a way to do it all at once as there are 24000 columns. And also how would I be able to combine the averaged data into 20sec intervals, I tried the buffer function but that didn't work. My code can be seen below.

Fs = 2048;                                             % Sampling Frequency 
segment = 0.25;                                        % Segments (250 msec)
Samples_fft = round (Fs * segment);                         % Sample Length

L_UTR_int_fft = buffer(L_UTR_Detr, Samples_fft); %this will create columns that contain 20 sec of data


% 4) Complete the Fast Fourier Transform

Fs= 2048; % Sampling Frequency
L= length(L_UTR_Cond); % Signal Length
t= mean(diff((L_UTR_Cond)); %Sampling Time
Fn= Fs/2 % Nyquist Frequency

L_UTR_fft= fft(L_UTR_int_fft) 

P2=abs(L_UTR_fft /L);
P1=P2(1:L/2+1);
P1(2:end-1) = 2*P1(2:end-1);
 

L_UTR_fft_final = fs*(0:(L1/2))/L1; % I saw this online, and I think it needs to be done

Answer 1

[dpb]

"...I cannot figure out a way to take the average of each col[o]umn without manually inputing it. "

Fs = 2048;                                             % Sampling Frequency 
segment = 0.25;                                        % Segments (250 msec)
Samples_fft = round (Fs * segment);                         % Sample Length

L_UTR_mean=mean(buffer(L_UTR_Detr, Samples_fft),2);     % time-domain average of columns
...

For averaging PSD, fft will operate by column; simply then use MATLAB array operations to operate over each column as if it were a vector then average the result as above.

See documentation for all functions--virtually all have the optional dim input paramter to tell them which dimension you wish to operate over if not the default by column.

Some, like std are tricky in that the dim input isn't the second so you may need to use a placeholder to get the desired argument.

[Ryan Chhiba]

I'm sorry i dont think i asked my question properly, what you did makes sense. Here is my process;

1) Segment the data into 250msec intervals

2) conduct a short time fourier transform

3) Calculate the Median Frequency of the power spectrum over the time domain

4) Then combine the data points into 20 second intervals to then average

I think I can do steps 1-3 but I cannot figure out how to do step 4

load('Subject 2_VTAS CD_JTTT_T1_Feb_23_2021_conditionedEMG.mat');
Time_data= Processed_EMG.EMG_Processed.Segmented_Data.BIO_data_GPSTime ;
emg1= Processed_EMG.EMG_Processed.Conditioned_EMG.DDE_SCM_L_Notched_QA  ;
L_SCM_Cond= emg1 


Fs= 2048; % Sampling Frequency
Fn = Fs/2; % Nyquist Frequency
Fc= 6; % Lowpass filter (6 Hz)
t= mean(diff((L_SCM_Cond))); %Sampling Time

L_SCM_MVC= Segmented_trackingData.biosemi.emg_MVC.MVC_emgMAX.maxabs_EMG_SCM_L;
% Remove the bias

L_UTR_Detr = detrend (L_SCM_Cond); % to remove the DC offset


% 1) Segment data into 250 msec intervals 

Fs = 2048;                                             % Sampling Frequency 
segment = 0.25;                                        % Segments (250 msec)
Samples_fft = round (Fs * segment);                         % Sample Length

% 2) Fourier Transform
L_UTR_int_fft = buffer(L_UTR_Detr, Samples_fft); %this will create columns that contain 250 msec of data
Fs= 2048; % Sampling Frequency
L= length(Time_data) 
t= mean(diff((L_SCM_Cond))); %Sampling Time
Fn= Fs/2; % Nyquist Frequency

L_UTR_fft= fft(L_UTR_int_fft);

P2=abs(L_UTR_fft /L);
P1=P2(1:L/2+1);
P1(2:end-1) = 2*P1(2:end-1);
 
figure(3)
f= Fs*(0:(L/2))/L;
plot(f,P1) % this is the plot of all the frequencies 

%3) for medain frequency for power spectrum athrough time series
L_UTR_Medfreqemg= medfreq(L_UTR_fft); 

[dpb]

Well, I don't know what you think Step 4 means, precisely. Once you've done the FFT, you've thrown away time.

If you want ensemble averages over 20-sec intervals instead, why not just compute them?

[Ryan Chhiba]

What I need is the the Median Frequency values of this raw emg data in 20 second intervals to compare it to the root means squared data I have of theis same data over 20 second intervals.

• From my understanding, in order to accomplish that I need to conduct a short time fourier transform and then get the median frequency for those intervals. Then I have to find a way to dicplay it as averages over 20 second intervals

"If you want ensemble averages over 20-sec intervals instead, why not just compute them? " - great point, but I do not how to do that

[dpb]

Same way as for 250 msec intervals???

[Ryan Chhiba]

I understand, but in order for the data to be accurate, a short time fourier transform must be conducted first. This requires an FFT over small epochs (250ms) to help better identify trends. This is why they need to be broken into small epochs, After caluclating the medfreq, I tried to use the buffer function on the data again but it would not work.