The PianoSens sensor is giving so much more information than the mic. The sensor is able to grab up to 10 partials without hitting a null / node. It accomplishes this by positioning between the hammer and the agraph. The hammer is naturally 1/8th the string length so it nulls out the 8th partial acoustically. So by positioning it closer to the agraph, we are 'guaranteed' to pick up the 8th, 9th, and 10th partials without sitting on a node.
I presented my class to the San Diego PTG. They came up to the piano and tried out the sensor, and switched back and forth from the mic to the sensor. We used two apps: Cybertuner and Pianoscope. They experienced first hand as 1 cent differences, worst case,(sometimes the same). It was inconsistent with the mic. I had shown examples of this in my class presentation first, then I wanted them to 'experience' it themselves.
I used to think the reduced 'jitter' or variance of the ETD's indicator was the most important feature of the sensor, but now I believe it is the increased accuracy of the frequency detection.
Best regards,
'Profesor' Steve
------------------------------
Steven Norsworthy
Cardiff By The Sea CA
(619) 964-0101
------------------------------
Original Message:
Sent: 01-07-2024 18:29
From: Steven Norsworthy
Subject: Why mic positions cause inaccuracy in frequency readings in your ETD's
People ask Professor Steven (me) why when we change the mic or change its position, we get frequency errors and variations. They say, "I understand why I get amplitude variances, but not frequency variance." Here is why. Every signal processing engineer (my peers) know that when we drop down in amplitude closer to the noise floor, we loose signal-to-noise ratio (SNR) but also get frequency error estimates. What is happening is that the partials we need to detect are too low and 'smear' in frequency and get pertubated in frequency by the noise as well. What we want are very well defined sharp spectral lines of the partials. The underlying FFT (Fast Fourier Transform) of the ETD has a window function that widens (by necessity) that spectral line, and we take the peak of that lobe through interpolation to estimate that frequency more precisely. Some ETD's do that better than others, by the way. When we use a 'rectangular window' FFT, we only have about 13 dB SNR to play with. Below that, we cannot detect the peak of the lobe very well and hence we are off the frequency in error. Hence, we get a noisy reading in frequency in our ETD. The PianoSens device eliminates this because it it picking up only string vibrations, not mic placement errors coming from nodes (nulls) from interference patterns.
Kind and eager to teach signal processing to the PTG community.
Steven Norsworthy
Professional Signal Processing Engineer, Lecturer, Inventor
Professional Musician, Pianist
PianoSens.com
RF2BITS.com
------------------------------
Steven Norsworthy
Cardiff By The Sea CA
(619) 964-0101
------------------------------