Feature request: Subtract trace in decibels

Antti Huovilainen

New Member
Thread Starter
Joined
May 15, 2017
Messages
10
Hi,

Another low hanging fruit:
Add feature to calculate dB(A) - dB(B). This should be trivial to implement and would be useful to separate the effects of comb filtering / modal ringing (can't be compensated with EQ above 100 Hz) from general tonality (can largely be compensated with EQ).
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,198
That's largely what A/B produces, albeit taking phase into account. Can you give some example data illustrating the use you have in mind?
 

Antti Huovilainen

New Member
Thread Starter
Joined
May 15, 2017
Messages
10
In that case, it appears that A/B has a bug that makes it unusable if the first response is original but the second response has been generated by averaging. In that case the resulting response is at -240 dB for all points (or a NaN is interpreted as -240 dB).

My use cases are comparing the difference between L & R (to see if the tonality differs too much) and for visualizing the amount of ripple vs tonality (subtract third octave averaged response from the original) .
 

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,198
In that case, it appears that A/B has a bug that makes it unusable if the first response is original but the second response has been generated by averaging. In that case the resulting response is at -240 dB for all points (or a NaN is interpreted as -240 dB).
Behaves as expected for me, no odd results. Can you post an mdat file with data that shows that problem?
 

Antti Huovilainen

New Member
Thread Starter
Joined
May 15, 2017
Messages
10
Here's an mdat which shows the problem.
I applied Psychoacoustic smoothing to L1, pressed "Average the Responses" (with only L1 selected), then chose the "A / B" trace arithmetic with L1 as A and Average as B. Resulting response is -240 dB at all points.
 

Attachments

  • rew_trace_bug.zip
    1.9 MB · Views: 18

John Mulcahy

REW Author
Joined
Apr 3, 2017
Messages
7,198
Thanks, problem occurs when the B trace is 96 PPO log spaced and the A trace is linear spaced. If you swap the order of A and B it will work OK, or turn off the option to allow 96 PPO log spacing in the analysis preferences before producing the average. Either highlights a different bug though in that the ratio of a smoothed measurement to the magnitude-only average derived from it isn't the 0 dB flat line it should be. A workaround for that is to use a vector average (or a vector sum which is easier for multiple measurements) instead. I'll address both issues at the weekend.
 
Top Bottom