What is the purpose of millidecades vs decidecades in software MANTA?

Question

The new software called MANTA makes power spectral density plots for various percentiles for any file or set of files you put into the GUI.

Bitbucket site for the software is here: https://bitbucket.org/CLO-BRP/manta-wiki/wiki/Home

When I look at the large excel file that it outputs, though I notice that it has a column for 1-Hz band bins up to 400 Hz, and then switches to millidecade bands. I am used to decidecade bands. The standards for modeling and measuring soundscape data as set by the ADEON project (https://adeon.unh.edu/sites/default/files/user-uploads/WEB-DRAFT%20ADEON%20Soundscape%20Specification_V2.pdf) define it as “one tenth of a decade (ISO 18405). Its value is approximately equal to that of one third of an octave, and for this reason is sometimes referred to as a ‘one‐third octave’.” I’m aware that these bands are based on the perception of how we hear as humans.

The ADEON standards also suggest that “individual (unweighted) decidecade levels shall be stored in such a way as to permit appropriate frequency weighting in the future as understanding of animal hearing improves.” So I expected MANTA software to output decidecade band power levels in the resulting excel sheet. (Decidecade bands are covered nicely in Wikipedia at: https://en.wikipedia.org/wiki/One-third_octave)

However, MANTA stores millidecade bands. I can extract sets of columns following this table of decidecade band limits and then incoherently sum them (i.e., convert their dB values to microPascals, sum those, and convert the answers back to dB), but that is a lot of extra steps.

So, I’m curious to know why millidecade bands are needed in MANTA software instead of decidecade bands. Are they more important? Are they an up-and-coming standard for soundscape metrics? Please clarify.

Answer 1

That's a great paper! After an email conversation with the writers of MANTA, this is the answer they gave:

There are times when you need higher frequency resolution that is obtained using decidecade bands, and conventionally one then uses a Fourier transform with a 1 Hz band or similar. The benefit of a millidecade band over a decidecade band is precisely that higher resolution, so instead one could ask the question “what is the benefit of a millidecade band over a 1 Hz band?” The answer to that alternative question is that millidecade bands achieve a high resolution at high frequency without the huge storage overhead associated with a 1 Hz band FFT up to (say) 100 kHz.

Check out work on millidecade bands by Bruce Martin and Michael Ainslie and others for more literature.

Links to such literature include:

Wall et al., 2021 https://www.frontiersin.org/articles/10.3389/fmars.2021.703682/full

Dekeling et al., 2014 https://www.semanticscholar.org/paper/JOINT-WORKSHOP-REPORT%3A-PREDICTING-SOUND-SOUNDSCAPE-Dekeling-Erbe/0954da435eef06740c58945b9c14ff4ad57755ea

and Page 64 in the book "Innovation and Discoveries in Marine Soundscape Research" https://books.google.com/books?id=_JhnEAAAQBAJ&pg=PA64&lpg=PA64&dq=LTSAs+and+millidecade+bands&source=bl&ots=WOkBU0XGKG&sig=ACfU3U0-wTM0E1UHOMQeSjIj3J9ifxOSFA&hl=en&sa=X&ved=2ahUKEwigz8zzqef4AhXWIjQIHTUWCHwQ6AF6BAgYEAM#v=onepage&q=LTSAs%20and%20millidecade%20bands&f=false

I was surprised to find that the following books do not have "third-octave bands", "decidecade", nor "millidecade" in there indices:

Urick - Principles of Underwater Sound

Kay - Fundamentals of Statistical Signal Processing Detection Theory

Richardson et al - Marine Mammals and Noise

Ainslie - Principles of Sonar Performance Modeling

Morse and Ingard - Theoretical Acoustics

Answer 2

I think this JASA Express Letter would help answer your question. In it, the authors (which include all of the MANTA leads as listed on the MANTA wiki) advocate for the use of hybrid millidecade spectra for use in long-term soundscape data storage and exchange. To summarize, as I understood their article, they felt that the hybrid millidecade metric was of sufficient resolution for a wide variety of applications, more than a decidecade spectra, while still keeping the size of data files manageable. As you mention, you can calculate decidecade bands or other coarser resolution sizes from millidecade bands if need be, but you wouldn't be able to get a finer resolution from decidecade data if you needed to, so it also increases the data's reusability.

I hope this helps!

S. Bruce Martin, Briand J. Gaudet, Holger Klinck, Peter J. Dugan, Jennifer L. Miksis-Olds, David K. Mellinger, David A. Mann, Olaf Boebel, Colleen C. Wilson, Dimitri W. Ponirakis, and Hilary Moors-Murphy , "Hybrid millidecade spectra: A practical format for exchange of long-term ambient sound data", JASA Express Letters 1, 011203 (2021) https://doi.org/10.1121/10.0003324