I some time ago I have started to use the Konno-Ohmachi smoothing function for smoothing of my spectra and calculating SSR). However, recently I have checked the results and it turnes out that during normal application, the smoothened spectrum is rather deformed when it came to amplitude – in came cases ridiculously high, in others rather low. The problem can be partially fixed by turning on the normalization, but the results are still different than I would expect (in general higher for low frequencies, lower for high frequencies). Is there any catch in the application of the smoothing?
I am sorry if the answer is obvious for a seismologist – I am an engineering geologist, learning seismology sort of on the way.
Follows my application of the script (just in case someone is interested in details):
from obspy import signal
from numpy import float64
import numpy as np
import obspy as ob
sample_rate = int(trace.stats.sampling_rate)
xF = np.fft.fft(trace.data)
N = len(xF)
xF = xF[0:N/2]
freq_axis = np.linspace(0,sample_rate/2,N/2)
freq_axis = float64(freq_axis)
frequency_spectra = float64(frequency_spectra)
frequency_spectra_smooth = signal.konnoohmachismoothing.konnoOhmachiSmoothing(frequency_spectra, freq_axis, bandwidth=100, max_memory_usage=1024, normalize=True)