Here's a pair of measurements, the first is with no dust covers. I'm using a Beringer measurement mic, and a scarlet 2i2 audio interface with 48khz sample rate. My mice is on the centerline of the panels, and about 1 meter away (slightly different between no dust cover and finished with grille cloth etc, sorry about that). The speaker base is about 20 inches off the ground on a small stand. Blue is no dust covers etc, and red is with dust covers, grilles (not usa, but normal aluminum expanded metal quad grilles), and grille cloth.
Sheldon
Is it possible to time align them and take A - B ?
I have to do this measurement again.... I remember playing square waves and measure with oscilloscope 25 years ago...
I have to do this measurement again.... I remember playing square waves and measure with oscilloscope 25 years ago...
@esl 63 @TNT Maybe first y'all can explain what we are looking at? In other words how to read the graphs. What should the "perfect" step response look like?
What is good and what is bad in these graphs? The less sharp points, the better? At what point the spikes occur? Difference in height between the top and the bottom of a spike?
Can you point out the prove of the advantages (or disadvantages in that matter) of leaving off the dust covers, grilles and speaker cloth shown in the blue graph compared to the red with some explanation?
Some related questions that come to mind:
Is the frame of the speaker not less rigid without the dust covers and the grilles taped in place? And how does THAT affect the graph?
Didn't PW calculate with the (extra) damping/impact/change by the dust covers, the grilles and the cloth, as it was part of the design?
What is good and what is bad in these graphs? The less sharp points, the better? At what point the spikes occur? Difference in height between the top and the bottom of a spike?
Can you point out the prove of the advantages (or disadvantages in that matter) of leaving off the dust covers, grilles and speaker cloth shown in the blue graph compared to the red with some explanation?
Some related questions that come to mind:
Is the frame of the speaker not less rigid without the dust covers and the grilles taped in place? And how does THAT affect the graph?
Didn't PW calculate with the (extra) damping/impact/change by the dust covers, the grilles and the cloth, as it was part of the design?
Like a.... step... as in a stair. But it is impossible as a spear can not uphold a DC (constant) pressure.
An impulse response should be a thin pin with no "foot". Also impossible due to that a membrane have a mass etc "problems".
//
An impulse response should be a thin pin with no "foot". Also impossible due to that a membrane have a mass etc "problems".
//
How fast confusion can kick in. The triangle is what I was expecting, so that is why I didn't understand your stair step.
OK, so now we got that out of the way, can we go back to the interpretation of the graphs please. The differences.
OK, so now we got that out of the way, can we go back to the interpretation of the graphs please. The differences.
No. The step response of a bandpass is NOT a triangle. Sheldon's graphs show near-perfect responses.
For illustration, two graphs of a mixed-phase step response of a linear system, red graph, and the one of a bandpass approximating more or less a loudspeaker response (Butterworth bandpass 4.Order 50Hz ... 18kHz), brown graph. Any bandpass will show it's specific impulse or step response. But never a geometrically clean triangle.
This one is cut 2ms before, and 7ms after the impulse, approximating Sheldon's graph:
Coming back to Sheldon's graphs, the red graph of the fully fitted ESL seems closer to a step response of a clean bandpass than the blue one of the naked system. I'd whonder how the two variants would differ in a wavelet transform.
And remember ... in practice there is no "ideal" step response. Each step response corresponds to a distinct response in frequency range. So in transformation, speaking of an "ideal" step response is like speaking of an "ideal" frequency response.
For illustration, two graphs of a mixed-phase step response of a linear system, red graph, and the one of a bandpass approximating more or less a loudspeaker response (Butterworth bandpass 4.Order 50Hz ... 18kHz), brown graph. Any bandpass will show it's specific impulse or step response. But never a geometrically clean triangle.
This one is cut 2ms before, and 7ms after the impulse, approximating Sheldon's graph:
Coming back to Sheldon's graphs, the red graph of the fully fitted ESL seems closer to a step response of a clean bandpass than the blue one of the naked system. I'd whonder how the two variants would differ in a wavelet transform.
And remember ... in practice there is no "ideal" step response. Each step response corresponds to a distinct response in frequency range. So in transformation, speaking of an "ideal" step response is like speaking of an "ideal" frequency response.
Last edited:
This part was what I called the triangle as a simple one word interpretation of the first part of the graph.
Again confusion or wrong use of language on my behalf maybe.
Again confusion or wrong use of language on my behalf maybe.
This is a step response, which is not often used in speaker measurements. You normally use an impuls response. In theory, an impulse is an infinite amplitude and zero time 'spike' which is not possible (and not useful) in a physical system. The duration of the source impuls should be related to the bandwidth of the system you want to measure.
The system response to an impuls provides information about the system damping (how fast it returns to zero position); the faster it returns to equlibrium, the better damped it is.
The frequency of the damped wave is related to the system resonance frequency. Post #31 has some nice examples. The lower graph shows better damping than the upper assuming the X-scale is the same. The upper has a higher resonance frequency.
It should be noted that the return to equlibrium is never a straight line or triangle or whatever, it is a sinusoidal part of the vibrating system resonance frequency while it is returning to rest position.
With the right math, you can completely characterize the system from the impulse response.
Jan
Last edited:
Last graph was scaled to match the scaling of Sheldon's graphs. All step responses were the same, as stated, from a linear and from a 4th order Bandpass.
Have some examples with bandpasses (e.g. of different orders or different damping) to illustrate your point. The following graphs show bandpasses with the same corner frequencies 50Hz and 18kHz as before. Instead, different orders of 1st (green), 2nd (brown), 3rd (blue) and 4th ( grey) order were chosen. You may compare now e.g. the mixed phase step response for the 1st oder and the 4th order filter. Something similar would occur in the step response when tweaking the damping of the filters:
Last edited:
Yes. And I would add ... for best possible S/N ratio. Because testing with an impulse as the source comes along with a small stimulation energy compared to e.g. a sinesweep measurement. Which makes the impulse testing method quite vulnerable in terms of S/N ratio. Therefore and as a countermeasure, you may increase the duration of the test pulse.
Beware: Pulse testing with pulse lengths > 1 comes along with a low pass, comb filtering characteristic in frequency domain. While the response of f_min gets raised, the response at f_max remains 0dB. In the following graphs you can see that e.g. an impulse lenght of 4 (brown graph) expectedly comes along with a DC gain of 12dB and drops by -2dB @ 20kHz. In these examples, SR was chosen 192kHz.
Last edited:
@ Jan Didden: Yes, and probably since 0.xy - Acourate is a nice tool for filter simulations/studies also.
An important addendum to the multilength test impulses lowpass chraracteristic: The HF drop is linearly dependent from the sampling rate.
As mentionned, the -2dB @ 20kHz for a pulse lenght of 4 is valid for a SR of 192kHz. For a more conventional measuring SR of 48kHz the high-frequency falloff looks quite dramatic: For the pulse lenght of 4 you linearly get -2dB at 5kHz already. Because 192kHz/48kHz == 20kHz/5kHz:
For SR = 48kHz:
An important addendum to the multilength test impulses lowpass chraracteristic: The HF drop is linearly dependent from the sampling rate.
As mentionned, the -2dB @ 20kHz for a pulse lenght of 4 is valid for a SR of 192kHz. For a more conventional measuring SR of 48kHz the high-frequency falloff looks quite dramatic: For the pulse lenght of 4 you linearly get -2dB at 5kHz already. Because 192kHz/48kHz == 20kHz/5kHz:
For SR = 48kHz:
No one uses either. The stimuli is in almost 95% of the cases a sweep or maybe perhaps a chirp. But never a step or an impulse as actual stimuli to speaker terminals.
Step and impulse are derived from sweeps / FFT.
//
It has been pointed out to me that these days doing an impulse or step response is actually done with a frequency sweep or chirp, followed by FFT processing.
I'm getting old ...
Jan
I'm getting old ...
Jan
Never say never again.
Time ago Liberty Instrument IMP worked with a single impulse stimulus, with the option to average multiple runs in order to improve the S/N ratio by averaging the ambient noise. Sometimes one of these "averaging" pulses completely messed up the measurement, because this one single pulse had been overloaded by the neighbour's dog barking, or the neighbour himself slamming a door. Then came IMP/M with a MLS (minimum lenght sequence) hardware generator with a much higher stimulus energy and consequently with more robust and better S/N ratio results. And then came the sinesweep method along with capable soundcards, which made MLS/M obsolete.
Recently, the option for single pulse measurements has been implemented in Acourate. Because a single pulse measurement has it's very own merits. E.g. other than with the sinesweep approach, the single pulse will at any time stimulate an electrically and mechanically inactive DUT. A short, single pulse also has it's very distinct phase properties, so the DUT's assessment of it's phase behavior over time might be more precise than with the sinesweep method.
Welcome back to the future ...
Last edited: