Filtering noisy data

[Alyse Kehler]

Hi,
I am having trouble completely filtering out some force data, because I believe the noise is at the same frequency as the data I need.
So far I have been using a butterworth 4th order filter like so. This works for running and walking force data, but now I am working with roller skiing.

sampling_freq = 1000;
cutoff =20;
[B,A]=butter(4,cutoff/(sampling_freq/2),'low');

and the traces come out with an obvious sine wave and extremely noisy.
So then I figured out what frequencies of noise there when during a zero trial on the treadmill using a fast fourier command:

function spectral_peaks=fast_fourier(data,Fs)
% inputs:
% data: column vectors with frequencies to be analyzed
% Fs: input frequency data was recorded at

Fs=Fs;
m=length(data);
n=pow2(nextpow2(m));
y=fft(data,n);
f=(0:n-1)*(Fs/n);
power=y.*conj(y)/n;
plot(f,power)
xlabel('Frequency (Hz)')
ylabel('Power')
title('Periodogram')

signals=findpeaks(power);
for n = 1:length(signals)
frequency_index(n,:)=find(power==signals(n));
end
spectral_peaks=f(frequency_index(:,1));

And then I would use a bandstop filter to remove all those frequencies. But the problem I am running into is that first, there are a lot of noise frequencies to remove and I am simultaneously removing most of my data as well.

Any suggestions or other ways I could clean up my plots while still retaining my data?
Thanks so much, Alyse