Hi everyone,
While reading about FSAF (Free Soundfield Analysis), I recalled that an Italian hi-fi magazine (Audioreview) developed a novel measurement technique years ago. Their goal was to improve the correlation between loudspeaker measurements and perceived sound quality during listening sessions.
After tracking down the original paper, I searched online for references and discovered—to my surprise—that they’ve republished all three technical articles for free.
Since the articles are in Italian, and because they include a free Java-based tool (created by a reader) to measure TND, I used ChatGPT and others AI tools to help me to summarize and translate their content.
Below you’ll find the summary and a link to download the tool (hopefully not too outdated!).
Note: After the publication of the first two introductory articles, Audioreview added this measurement technique to its standard loudspeaker test protocol.
Important Addendum:
I found a 2009 AES paper that references this technique. The article is titled "Silence Sweep: A Novel Method for Measuring Electro-Acoustical Devices" by Angelo Farina, Department of Industrial Engineering, University of Parma.
Although I found a link to the publication, I wasn't able to download it directly.
However, it is accessible via the Wayback Machine at the following address:
http://pcfarina.eng.unipr.it/Public/Papers/247-AES126.pdf
Enjoy!
References:
https://www.audioreview.it/tecnica/total-noise-distortion.html
https://www.audioreview.it/tecnica/total-noise-distortion-seconda-parte.html
https://www.audioreview.it/tecnica/total-noise-distortion-per-tutti.html
Software Tool Download:
http://audioreview.it/wp-content/uploads/download/dist-last.zip
Credits & References to the series of articles originally published in Audioreview and recently republished online:
AI generated summary:
1. Technical Summary of the Audioreview TND Articles
**a. Introduction of the TND Concept (from AUDIOreview no. 268 May 2006)
Montanucci et al. introduce Total Noise Distortion (TND) as a broadband measure of all non‑linear artifacts (harmonic + intermodulation + “masking” distortion) generated by a loudspeaker under a continuous‑spectrum stimulus. Instead of single‑tone THD or dual‑tone IMD tests, TND uses complementary signals (“A” and “B”) each containing the full spectrum except for alternating narrow “holes” (third‑octave bands), and by measuring the energy that “fills” those holes across the two phases, it directly captures distortion at the same frequencies as the stimulus—automatically compensating for room reflections and eliminating the need for anechoic conditions.
Key points:
The second article confirms that TND is environment‑independent (anechoic vs. reflective shows only minor deviation) and stimulus‑agnostic: the methodology applies to any continuous‑spectrum source, including music. By splitting a music track into complementary spectral bands (A/B) just as with noise, one can measure the real‑world non‑linear coloration introduced when playing actual program material—fully in situ, without special acoustical treatment.
c. Free Java Software for TND (“for everyone”, from AUDIOreview no. 288 March 2008)
Fabrizio de Leonardis released a Java application that automates:
From the article photos (included translated original comment) below you can see the influence of a high-pass filter on the measured distortion:
Figure 7. Total Noise Distortion of the Quad 989 system at 90 dB pressure. Although the distortion is low overall, and very low in the mid and high range, there is a lot of “lateral banding”, with high peaks at the band boundaries. This is all induced by the movement associated with the lowest component of the signal, because if we filter the test signal at 500 Hz we see a real collapse of the residues. A much more marginal reduction is observed if the cutoff is further raised to 2000 Hz.
Figure 8. Klipsch Synergy B2, measurement taken at 90 dB. If in this case we cut at 500 Hz, the TND residues produced by the woofer in the mid-range drop to “super” transducer levels, and the peaks on the side bands disappear completely.
Figure 9. Library Another small two-way, Linea Indiana HC205. Here we note that to obtain an already drastic improvement in the behavior of the low range, a cut at 100 Hz is enough. In cases (frequent, from what we have observed so far) like this, the more the woofer is forced to work in the low range, the more the TND in the mid range increases.
Cheers,
Roland
While reading about FSAF (Free Soundfield Analysis), I recalled that an Italian hi-fi magazine (Audioreview) developed a novel measurement technique years ago. Their goal was to improve the correlation between loudspeaker measurements and perceived sound quality during listening sessions.
After tracking down the original paper, I searched online for references and discovered—to my surprise—that they’ve republished all three technical articles for free.
Since the articles are in Italian, and because they include a free Java-based tool (created by a reader) to measure TND, I used ChatGPT and others AI tools to help me to summarize and translate their content.
Below you’ll find the summary and a link to download the tool (hopefully not too outdated!).
Note: After the publication of the first two introductory articles, Audioreview added this measurement technique to its standard loudspeaker test protocol.
Important Addendum:
I found a 2009 AES paper that references this technique. The article is titled "Silence Sweep: A Novel Method for Measuring Electro-Acoustical Devices" by Angelo Farina, Department of Industrial Engineering, University of Parma.
Although I found a link to the publication, I wasn't able to download it directly.
However, it is accessible via the Wayback Machine at the following address:
http://pcfarina.eng.unipr.it/Public/Papers/247-AES126.pdf
Enjoy!
References:
https://www.audioreview.it/tecnica/total-noise-distortion.html
https://www.audioreview.it/tecnica/total-noise-distortion-seconda-parte.html
https://www.audioreview.it/tecnica/total-noise-distortion-per-tutti.html
Software Tool Download:
http://audioreview.it/wp-content/uploads/download/dist-last.zip
Credits & References to the series of articles originally published in Audioreview and recently republished online:
- The original technical solution was conceived by Fabrizio Montanucci, Technical Editor-in-Chief of Audioreview, who also curated the related article.
- The theoretical framework is based on the work of Prof. Francesco Romani, Professor of Algorithms and Data Structures at the University of Pisa.
- The Java software implementing the concepts presented in the previous articles was developed by Fabrizio de Leonardis, a long-time Audioreview reader and co-author of the third and final article in the series, which introduces and documents the software.
AI generated summary:
1. Technical Summary of the Audioreview TND Articles
**a. Introduction of the TND Concept (from AUDIOreview no. 268 May 2006)
Montanucci et al. introduce Total Noise Distortion (TND) as a broadband measure of all non‑linear artifacts (harmonic + intermodulation + “masking” distortion) generated by a loudspeaker under a continuous‑spectrum stimulus. Instead of single‑tone THD or dual‑tone IMD tests, TND uses complementary signals (“A” and “B”) each containing the full spectrum except for alternating narrow “holes” (third‑octave bands), and by measuring the energy that “fills” those holes across the two phases, it directly captures distortion at the same frequencies as the stimulus—automatically compensating for room reflections and eliminating the need for anechoic conditions.
Key points:
- Stimulus generation: multichannel FFT carving of complementary “holes” into pink noise, achieving ≈0.01 Hz transition steepness and low crest factor (~9.8 dB), enabling high‑level testing.
- Measurement: play signal A, record; play complementary signal B, record; combine residual spectra to form a single TND curve (% vs. frequency).
- Insights: As the number of simultaneous tones increases, measured harmonic distortion drops but total distortion grows—revealing intermodulation effects invisible to single‑tone tests.
The second article confirms that TND is environment‑independent (anechoic vs. reflective shows only minor deviation) and stimulus‑agnostic: the methodology applies to any continuous‑spectrum source, including music. By splitting a music track into complementary spectral bands (A/B) just as with noise, one can measure the real‑world non‑linear coloration introduced when playing actual program material—fully in situ, without special acoustical treatment.
c. Free Java Software for TND (“for everyone”, from AUDIOreview no. 288 March 2008)
Fabrizio de Leonardis released a Java application that automates:
- Low‑crest‑factor pink‑noise generation (9.7–9.8 dB) via Romani’s fragment‑assembly algorithm.
- Complementary file creation through FFT, band nulling, and inverse FFT; user‑selectable band ranges, equally applicable to music tracks.
- Batch TND measurement on recorded .wav pairs (1 M–4 M samples, 16/24 bit) with built‑in plotting.
- Window weighting and transition‑band rejection to eliminate spurious artifacts.
Limitations include a 16‑bit noise floor (~0.01 % distortion) and ≥2 GB RAM for large sample runs.
- Noise Sensitivity : Environmental and hardware noise limit accuracy.
- Low-Frequency Resolution : Requires long measurement times (e.g., 24+ seconds per signal) for sub-100 Hz analysis.
- Hardware Dependency : Lower-quality gear (e.g., budget mics) introduces errors.
From the article photos (included translated original comment) below you can see the influence of a high-pass filter on the measured distortion:
Figure 7. Total Noise Distortion of the Quad 989 system at 90 dB pressure. Although the distortion is low overall, and very low in the mid and high range, there is a lot of “lateral banding”, with high peaks at the band boundaries. This is all induced by the movement associated with the lowest component of the signal, because if we filter the test signal at 500 Hz we see a real collapse of the residues. A much more marginal reduction is observed if the cutoff is further raised to 2000 Hz.
Figure 8. Klipsch Synergy B2, measurement taken at 90 dB. If in this case we cut at 500 Hz, the TND residues produced by the woofer in the mid-range drop to “super” transducer levels, and the peaks on the side bands disappear completely.
Figure 9. Library Another small two-way, Linea Indiana HC205. Here we note that to obtain an already drastic improvement in the behavior of the low range, a cut at 100 Hz is enough. In cases (frequent, from what we have observed so far) like this, the more the woofer is forced to work in the low range, the more the TND in the mid range increases.
Cheers,
Roland
Last edited:
That's quite interesting - from the copy at archive.org:
https://web.archive.org/web/20160817105637/http://pcfarina.eng.unipr.it/Public/Papers/247-AES126.pdf
https://web.archive.org/web/20160817105637/http://pcfarina.eng.unipr.it/Public/Papers/247-AES126.pdf
Okey - so you listen in a band where there should be silence and record what's going in there.... interesting!
//
//