Harmonic analysis Equalize2

[[PTMAO]Dusk]

I was quite surprised to watch some noise (under -180dBFS) in plugindoctor when comparing to other eqs .

[martinjuenke]

How does it sound? Good?

[[PTMAO]Dusk]

Acon equalize is an excelent tool. i love it. I just did a quick test on plugin doctor and noticed that. just wondering why. nothing else

[fese]

Just a wild guess, as I am no DSP developer:
Could it be truncation distortion? All these EQs will do their internal calculations in 64 bit, but maybe Equalize2 doesn’t support the doubleProcessing method (or whatever it is called) of the VST spec, so the 64bit get truncated to the 32bit output. Just a wild guess.
Anyhoo, as the noise is way below that of the analog components of your converters, it won’t really matter.

[stian]

Please be aware that 200 dB range is a very extreme range and that the distortion levels here will never be audible. A 24 bit audio file has 144 dB dynamic range, so the truncation to 24 bit will mask this noise anyway. The technical reason for the distortion is that Equalize uses 32 bit FIR filtering. Many equalizers use IIR filters where 64 bit is a definite requirement because noise builds up in the feedback. This is not the case with FIR filters, where 32 bit is sufficient for high end audio.

Best,
Stian

[mystran]

Please be aware that 200 dB range is a very extreme range and that the distortion levels here will never be audible. A 24 bit audio file has 144 dB dynamic range, so the truncation to 24 bit will mask this noise anyway. The technical reason for the distortion is that Equalize uses 32 bit FIR filtering. Many equalizers use IIR filters where 64 bit is a definite requirement because noise builds up in the feedback. This is not the case with FIR filters, where 32 bit is sufficient for high end audio.

Yeah, what we see here is just the normal quantization noise of 32-bit signals. It doesn't even take any processing whatsoever, just convert any 64-bit signal to 32-bits before sending to FFT and it'll show roughly similar noise floor... yet it's still way more accurate than anything any physically plausible DAC will ever be able to output.

edit: Also not all IIR filters are created equal. Direct forms are notoriously bad, but implicit trapezoidal filters (better known as ZDF filters) can be perfectly usable in 32-bit as well (with slightly higher noise floor than what's visible here, but not so much that it'd make any practical difference).

[[PTMAO]Dusk]

Thanks for all replies. I´m a wiser person now eheh

[stian]

edit: Also not all IIR filters are created equal. Direct forms are notoriously bad, but implicit trapezoidal filters (better known as ZDF filters) can be perfectly usable in 32-bit as well (with slightly higher noise floor than what's visible here, but not so much that it'd make any practical difference).

Thanks, your're right. I have only experimented with bi-quad structures, but there are surely better topologies to avoid noise accumulation.

Best,
Stian

[rafa1981]

Not that I understand all that, but leaving the papers for the trapezoidal SVF:
https://cytomic.com/index.php?q=technical-papers

[stian]

Not that I understand all that, but leaving the papers for the trapezoidal SVF:
https://cytomic.com/index.php?q=technical-papers

Thanks for the links! I'm afraid that these filter topologies don't give us the flexibility we need for Equalize, though.

Best,
Stian