How to prevent oversmoothing while applying fft to raw signal

Question

I am trying to do fft for removing noise from the signal. While doing that I am getting frequency domain like thisfreq_domain:

So when applying butter pass filter to the peak freq of the signal is over smoothing like this original image:

image after applying butter pass filter:

So I am stuck at this and want to figure out the solution for reducing over smoothing in the signal

for i in range(data_first_interval.shape[0]):         
       ppgwave=data_first_interval.loc[i]        
       ppg_fit=fftpack.fft(np.array(ppgwave))        
       ppgarr=np.array(ppgwave)        
       amp=2 / time_vec.size*np.abs(ppg_fit)        
       sample_freq=fftpack.fftfreq(2100,0.001)        
       signal_amplitude = pd.Series(amp).nlargest(2).round(0).astype(int).tolist()    
       magnitudes = abs(ppg_fit[np.where(sample_freq >= 0)])    
       #Get index of top 2 frequencies\    
       peak_frequency = np.sort((np.argpartition(magnitudes, -2)[-2:])/2.1)    
       cutoff = peak_frequency[1]    
       y = butter_lowpass_filter(ppgarr, cutoff, fs, order)     
       data_first_interval.loc[i]=y

and my butter low pass filter is as defined below

fs = 1000.0           
order = 2          
def butter_lowpass_filter(data, cutoff, fs, order):    
   print("Cutoff freq " + str(cutoff))    
   nyq = 0.5 * fs # Nyquist Frequency    
   normal_cutoff = cutoff / nyq    
   # Get the filter coefficients     
   b, a = butter(order, normal_cutoff, btype='low', analog=False)    
   y = filtfilt(b, a,data)    
   return y 

Can anyone help me where I am going wrong

Answer 1

Typically the way you produce "less smoothing" is to raise the cutoff frequency.

With smoothing filters, you have two parameters to play with:

• Cutoff: higher cutoff produces less smoothing; lower cutoff produces more smoothing.

• Order (related to "roll off"): higher order causes less low frequency distortion, but may cause ringing. Lower order has less ringing, but may distort frequencies below the cutoff. This may also appear as over smoothing.