fast RMS estimator

exonerate · 20 Mar 2010

Thanks 🙂

So this is a 'fake' RMS using a lowpass filter?

How did you get the constant factor? I'd like to make that sample rate independent.

Cheers 😉

Exo

bootsy · 20 Mar 2010

no, this is a realtime continuous RMS and not a simple LP filter. At double SR the constant should be the half, I think ...

oddson · 20 Mar 2010

I'm not ASM savvy but I see a root and a square but where's the mean?

It seems like it is still doing something like a low-pass to achieve the averaging part since it looks like there's only 1 sample worth of memory here.

Can you write out what it's doing in pseudo-code so we non ASM people can see what it's doing?

MichaelBenjamin · 20 Mar 2010

i would guess its like this:

rms=(in*in-rms)*0.00225+rms; out=sqrt(rms);

oddson · 20 Mar 2010

Then it's the root of a low-pass of the squared input...

It has the advantage of being an accurate reflection of the power of the signal because it is the root of the squared inputs 'averaged' in a sense.

So the coef determines the cutoff for the lowpass. It could be brought out of the code and made an adjustable parameter to give an approximation of the 'windowing' used for the 'average'.

(The scare-quotes are there because I am not really sure whether the terms are valid where I'm using them.)

bootsy · 21 Mar 2010

MichaelBenjamin wrote: i would guess its like this:

rms=(in*in-rms)*0.00225+rms; out=sqrt(rms);

it is. the constant is tuned to match it to an (array based) averaging calculation over 512 samples.

Andrew J · 21 Mar 2010

At 44.1KHz this seems to be within 0.5% accuracy on smooth monophonic signals such as vocals. Accuracy is out to 3 or 4% on something like a drum bus. The error can occasionally get out as far as 7% error on a full bus mix.

That's pretty good for something that runs at 1/6th the cycles of true RMS.

sunsynth · 25 Mar 2010

Hi bootsy,

i just tested your ASM RMS:

Even if i only input a Denormal value of 1e-020 the CPU rises 1.5 % 🙁

Please have a 2nd look @ it because i need an optimized ASM right now!

MichaelBenjamin · 25 Mar 2010

when you put a denormal number in its no surprise that cpu goes up. why would you do that anyway?

sunsynth · 25 Mar 2010

😊 ...Michael and bootsy

...in the heat

i have a schematic here and it 'prevents denormals with 1-e020' and i overlooked that i usually set a 1e-009 Value 😮

Go back to my knitting 😉

bootsy · 29 Mar 2010

So, as attached this should be the samplerate independent version. @Michael: Even on my ancient PC it takes only around 0.2% cpu but I've noticed that with shortest timings there could be indeed some denorm issue which one should compensate for ...

exonerate · 28 Apr 2010

Hey bootsy, I'm I ok to add this to the Dev toolkit?

Cheers 😉

Exo

bootsy · 28 Apr 2010

surething.

platinumears · 5 Oct 2010

bootsy wrote: So, as attached this should be the samplerate independent version.

Thanks for this bootsy. I used it in FL4TT3RY: http://www.kvraudio.com/get/4640.html

Much appreciated! 😀

MDMA · 30 Jan 2012

Thanks alot mate, I was looking for something like this and this one is really good. Nice work!