how can i remove Baseline wander of an ECG signal by two-stage moving-average filter?

i have an ECG siganl

fs = 360Hz

the first and second-stage averaging window

lengths are set to be 1/3 and 2/3 of the length of the

input signal in samples.

Answer 1

[William Rose]

@Naor p,

At each point, you replace the original signal with the orignal signal minus the moving average of the signal in that neighborhood.

If x(k)=original signal, k=1..N, and h(k)=moving average filter, k=1..M (where M is odd and M<N),

then you can make a new signal y(k):

y(k)=x(k)-sum{j=-(M-1)/2...+(M-1)/2 of h(j)*x(k+j)}

You'll have to do something at the edges to avoid issues.

[William Rose]

@Naor p

load 100m.mat
fs=360;
t=(1:length(val))/fs;  %time vector
M=301; %odd, approximately one beat duration
h=ones(M,1)/M; %flat moving average
xx=conv(val,h,'same'); %convolve val with the moving average window
%Next: y = original signal minus its moving average
y=val-xx;
%Next: plot results
subplot(2,1,1); plot(t,val,'-b',t,xx,'-r'); 
title('Original'); legend('Original','Mov.Avg.'); grid on
subplot(2,1,2); plot(t,y,'-r');
title('Remove baseline wander'); xlabel('Time (s)'); grid on

M=width of moving average window. Try it with different values of M. The behavior of the moving average at the end of the record is not ideal, because conv() pads the original signal with zeros at the end before doing the convolution. You could try something else, but it would get a bit complicated, and it might not generalize well to other EKG examples.

[Np]

thank you for your quick answer

[William Rose]

@Naor p, You are welcome.

@Star Strider is correct (as always) that the usual way to remove baseline wander is with a "standard" high-pass filter. Subtracting the moving average IS a form of high-pass filtering, but it is not standard. I can't see Star Strider's answer now. It was a good answer.

[Star Strider]

@William Rose —

Thank you!

I removed it because it didn’t address the ‘moving average’ issue. (It exists plentifully elsewhere, since that’s my usual approach to these problems.)

Also, @Naor p needs to Accept your Answer because it correctly addreses the baseline filtering problem using the requested approach. (That’s the other reason I reomoved mine.) I’m voting for it in the interim.

.