My contribution to the MOSQITO repository #63
Conversation
… han sido tocados
… result, but it works and returns a value.
…the max_level function
…onversion (spectrum2dBA.py)
…l.py functions (they all work)
|
Hello Ignacio, We are going to merge your work into a new branch to change a few things so that your code becomes coherent with the rest of the library. It will then be merged into the master branch :) Thanks for your contribution ! We'll let you know when it's done |
There was a problem hiding this comment.
The current version of 61672-1 is 2013. Annex E of the standard provides an analytical expression for spectral A-weighting as a function of frequency. It would be faster, more accurate, and more straightforward, to include A-weighting as expressed in equation E.6 of 61672-1:2013 Annex E.
Additionally, and this carries more widely to any related function definitions, the 'preferred' nominal fractional octave frequency band values should not be used for signal processing purposes - the exact frequency bands should always be used, which are defined in IEC 61260-1:2014. Any calculations should be carried out using the exact frequency values and then (if required, although this is perhaps rarely necessary, except for visualisation or data storage purposes) mapped to the corresponding nominal values using a matching algorithm.
There was a problem hiding this comment.
Why is the code operating over a signal in a for loop? These operations can all be applied to the entire signal in a single operation without any loops. This is true even if the input comprises multiple signals, although this code doesn't seem to address that situation.
This also goes for all other functions using this loop kernel.
Moreover, if the signal is being operated on sample-by-sample (as indicated), it doesn't make sense to take a mean of a single value (line 34).
Statistical levels are typically calculated using exponentially-weighted amplitude (ie, using a time-weighting integration filter), which is then downsampled to the specified calculation interval. For example, take time signal x, apply Fast (tau=0.125 s) time-weighting RMS filter to give xFRMS, then resample down to (eg) 0.125 s intervals. Then convert to dB and take the max/min/percentile.
Even if exponential weighting is not desired, taking an RMS implies averaging over a series of samples, which this code doesn't seem to be applying.
There was a problem hiding this comment.
See 61672-1:2013 Annex E for analytical expression for C-weighting as a function of frequency (equation E.1). The constant C1000 needed in that equation turns out to be ~0.061904281992313
Hi Martin,
I'm Ignacio Gracía, student of the URJC Fuenlabrada, Madrid. Here you have the result of my contribution to the project.
Best regards.