one-third octave synthesis from autospectrum fft

[Philippe Martin <philippemar@xxxxxxxxx>]

Dear list,

For a project, I need to measure the equivalent countinuous sound pressure level (Leq) in one-third octave. Unfortunatelly, the measuring device, a Pulse from Bruel & Kjaer (the only device I can use for this project) is only able to measure an autospectrum.

So, I tried to caculate the one-third octave spectrum by summing the autospectrum values sorted in the respectives bandwith. The result is not bad except for an almost constant bias of 2-3 dB compared to a measure done with a one-third octave sonometer. The difference appear on each band as well as on the global value.

Both devices are measuring the rms values.

Does anyone get an idea from where the difference can come from, or know about a tested algorythm to perform such a task?

Many thanks in advance,

Philippe