Measure the tactile spectrum for tuning my shaker setup

[cnhoff]

Hi,

i recently upgraded my tactile setup and am in the midst of trying to optimize the frequency response and effects and I can tell you, there is much to learn.

I am searching for a more objective way to measure the frequency response than my butt though.

I came across audio interfaces and corresponding software in the forums, and I am familiar with those from my guitar hobby, but am wondering what you are using as the sensor?

Stick-on piezo pickups or other microphones come to mind, but I would be interested to hear about your experience.

Is the frequency response of those flat enough over the relevant (5)20-200Hz bandwidth? Calibration with white noise would be possible, but the signal and audio path would also need to be flat enough...

What about the dynamic range, i guess you have to be careful to not drive them into nonlinearity or damage?

Thanks

Chris

 

[cnhoff]

Despite no replies to this thread, here is a small summary of my progress so far for anyone who might be concerned.

I have used a Focusrite Audio Interface with a freeware spectrum analyzer software called Visual Analyzer and a dual-piezo, mass loaded probe described by blekenbleu here to measure the spectrum and EQ the signal to be fairly flat as a starting point for the effects optimization.

I have used both sine-waves generated by a web-based generator and wav-file containing broadband noise as the sourcing signals.

I have no easy way to calibrate the piezo-probe, so i understand the measured spectra might not reflect the real spectrum 100%, but i found, my butt-meter is quite in line with what is measured and i never expected to-the-dB accuracy anyway.

The probe was taped to my aluminium seat plate containing amongst an LFE the characterized Clark TST329. The signal was bandpassed by my NX1000D driving the TST, but no EQ was applied for this measurement

One can clearly see (and this is in line with the felt sensation)

- the bandpass filtering applied through the NX
- a strong peak around 65 Hz and 160 Hz respectively
- up to 20dB lower power from 100-150Hz

Taking the measured spectrum as a reference, i came up with a 5-tap EQ set in the NX1000D and the resulting spectrum looks like this

The spectrum is quite flat in the frequency range of interest, which was my goal. The felt sensation when frequency-sweeping the sinusoidal signal is much more uniform than before and pretty much in line with the measurement, although i am aware that my body is neither an objective nor linear sensor across frequency and power.

This was only the first try, but i am pretty satisfied with the results so far. Thanks to blekenbleu for the thread which basically pencils out the approach.

This EQ setting is not perfect, but seems to be a quite good basis for further optimization. Ultimately the felt sensation is the reference for me. I will do similar measurements and EQ settings for the LFE and the LFE and TST on the pedal plate and see, where i reach with that.

 

[blekenbleu]

I have no easy way to calibrate the piezo-probe

I have read, but not confirmed, that smart phones can do some vibration measurements.
Something so large as your Clark TST329
should be able to energize a large enough plate, unlike my exciters,
to get comparable readings from a mobile device app and your piezo apparatus. The phone may want
anchoring, e.g. using double-sided carpet tape,
to mitigate spurious rattling.

FWIW, REW is intended for constructing EQ correction curves e.g. for subwoofers
which should be directly applicable to tactile devices.

Given rigid and elastic materials employed (e.g. aluminum extrusions),
I expect resonance decays to be problematic; so-called waterfall plots are wanted.

I don't know why/how I missed your initial thread message.

Calibration with white noise would be possible

Calibration from wide band stimulus energy is possible
with appropriate digital signal processing,
such as used for software defined radio,
but signal-to-noise ratio is usually an issue.
A swept sinusoid "chirp" requires minimal filtering on the receive side,
mostly time synchronization with the sweep. Amplifiers and transducers
can deliver more intense signals one frequency at a time
more easily than over all frequencies at once.

By analogy, best quality document scans were traditionally made by drum scanners,
which capture a single intense dot at a time. Next best were line (e.g. flatbed) scanners.
Photograph full frame high quality document captures want long exposures,
which require rigid copy stands and bright lights
that were dauntingly hot before LEDs were perfected.

 

[blekenbleu]

my body is neither an objective nor linear sensor across frequency and power

A larger effect is pressure; human skin sensitivity to vibrations decreases
with increasing pressure. A light touch is wanted. On the other hand,
with most sim racing tactile feedback coming thru hands, thighs and sit bones,
one is more interested in real perception with significant pressure.
One could try making just perceptible vibration measurements at different frequencies,
then just perceptible incremental difference measurements.

 

[cnhoff]

Calibration from wide band stimulus energy is possible
with appropriate digital signal processing,
such as used for software defined radio,
but signal-to-noise ratio is usually an issue.
A swept sinusoid "chirp" requires minimal filtering on the receive side,
mostly time synchronization with the sweep. Amplifiers and transducers
can more deliver intense signals one frequency at a time
more easily than over all frequencies at once.

Thanks for the reply.

That is fully understood and also the reason why SNR/making full use of the analyzer’s dynamic range-wise you are better off using a fast sine sweep using a network analyzer to do e.g. a broadband characterization of a linear network like a filter compared to doing the same with a band-limited noise or multi-tone signal.

By the way, the probe seems to be not as sensitive at the relevant LFE-frequencies below 40Hz, giving me some headache with the measurement, but I will experiment with better coupling tomorrow. Mechanics is not my specialty though…

 

[blekenbleu]

the probe seems to be not as sensitive at the relevant LFE-frequencies below 40Hz

Nut epoxied to one side of paired piezo sandwich for mass loading,
with hot glue or double sided carpet tape
securing the other piezo side to the vibrating surface.

My first few sandwiched piezo sensor attempts gave less consistent responses than later ones;
getting/keeping piezos parallel while epoxy cures is confounded by solder dots.
Dayton exciters are not interesting below about 25Hz;
I did not much explore optimal mass tuning of piezo responses at lower frequencies;
more than a single steel 13mm nut seemed not much better...

REW facilitates averaging results from multiple sweeps,
after rejecting sweep results confounded e.g. by spurious rattling.
Reduce tactile transducer drive to just above levels
where piezo 50/60Hz power noise pickup is problematic.
Differences in microphone cable shielding and grounding
significantly impact interference pickup.
If your Focusrite Audio Interface has a high impedance input option,
e.g. by 1/4-inch TRS jack for instruments, instead of low impedance by XLR,
definitely use the high impedance option.

Even for parts seemingly securely bolted together, measurements were cleaner
when double-sided tape was sandwiched between parts before bolting.

 

[cnhoff]

Nut epoxied to one side of paired piezo sandwich for mass loading,
with hot glue or double sided carpet tape
securing the other piezo side to the vibrating surface.
My first few sandwiched piezo sensor attempts gave less consistent responses than later ones;

getting/keeping piezos parallel while epoxy cures is confounded by solder dots.
Dayton exciters are not interesting below about 25Hz;
I did not much explore optimal mass tuning of piezo responses at lower frequencies;
more than a single steel 13mm nut seemed not much better...

Thanks. I think my main issues come from repeatability issues when the probe is not securely fixed to the seat or pedal plate. I have been using a strip of tape, but that can certainly be fine-tuned, I will try with double-sided tape this evening, as I do not want to use epoxy for that.

Based on your initial article I had already mass-loaded the dual Piezo-pickups with 5 nuts as it seemed you had best low frequency results with that setup. Also the two pickups are basically perfectly parallel to each other and I have used the high-impedance instrument setting.

I am using a cut 6,3mm (1/4” ?) instrument cable which is well shielded.

I read your initial thread quite thoroughly

REW facilitates averaging results from multiple sweeps,
after rejecting sweep results confounded e.g. by spurious rattling.
Reduce tactile transducer drive to just above levels
where piezo 50/60Hz power noise pickup is problematic.
Differences in microphone cable shielding and grounding
significantly impact interference pickup.
If your Focusrite Audio Interface has a high impedance input option,
e.g. by 1/4-inch TRS jack for instruments, instead of low impedance by XLR,
definitely use the high impedance option.

Even for parts seemingly securely bolted together, measurements were cleaner
when double-sided tape was sandwiched between parts before bolting.

The shown measurements were taken with 5 averages and occasionally max-hold (not in the posted ones) but obviously there must be a trade-off between averaging noise and measurement time.

The “automatic outlier-removal” sounds interesting, I will try REW later this evening.

 

[blekenbleu]

“automatic outlier-removal” sounds interesting

Outlier removal was not automatic;
I just accumulated trials until 5 or so had no outliers, deleted the rest, then averaged.
There is some integration available to pop REW's generated correction filters into Equalizer APO
for quick evaluation, even if they are eventually destined to be configured elsewhere.
REW also supports directly loading filters into a number of audio hardware devices.

 

[cnhoff]

Outlier removal was not automatic;
I just accumulated trials until 5 or so had no outliers, deleted the rest, then averaged.
There is some integration available to pop REW's generated correction filters into Equalizer APO
for quick evaluation, even if they are eventually destined to be configured elsewhere.
REW also supports directly loading filters into a number of audio hardware devices.

That sounds even more helpful, thanks.

I was thinking about exporting the data (time-domain or spectrum) to MATLAB for evaluation etc. but let’s see if that is even needed. Don’t want to write a paper, just want do reasonable measurements to tune my EQ in the end

 

[blekenbleu]

I will try with double-sided tape this evening, as I do not want to use epoxy for that.

I destroyed several piezo pairs by trying to remove after hot gluing..
Hot glue sometimes worked hardly at all and other times too well.

 

[blekenbleu]

MATLAB for evaluation

REW is quite a mature package with many home theater users..
including for waterfalls:

 

[blekenbleu]

I am using a cut 6,3mm (1/4” ?) instrument cable which is well shielded.

Do you carry ground thru the shield or via a separate wire?
Ideally, the shield is connected only at the jack,
making shielded twisted pair usually a better choice,
even though the piezo sandwich is not a well balanced input.
Among several old cables tried, some are much better than others for noise immunity.