That's how Genelec does it, S&R shows the newer Ones (8361A and 8331A) as phase linear starting from 1 kHz, the 8350A from 300 Hz and the S360 from 100 Hz.
Last edited:
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Step Response: Does It Really Matter?
- Thread starter watchnerd
- Start date
ernestcarl
Major Contributor
One must ALWAYS start with a decently designed physical situation.
But adjusting via DSP is no different than adjusting via moving drivers, in a very real fashion. The geometry is a bit different, but it's no better or worse.
But, absolutely, bad physical layout is, well, "bad physical layout". What I can I say?
- Joined
- Jan 15, 2020
- Messages
- 6,855
- Likes
- 16,780
I'm sorry, what? I'm not sure I qualify as "DIY" because I design loudspeakers (among other things) professionally, but I'm having some trouble decoding your point? Who mentioned just throwing DSP at something and ignoring the physics. It can be attempted, but nothing beats physics. Nothing.
- Joined
- Jan 15, 2020
- Messages
- 6,855
- Likes
- 16,780
bennybbbx
Addicted to Fun and Learning
ok i understand and learn, it is possible in theory with oversampling and interpolation(which can cause quality loss) to have any delay and not limit to sample steps. but i have no delay VST plugin that delay in other as sample steps. i search for subsample delay VST and find nothing. delay plugins allow feedback and allow easy to delay only 1 channel. did you know a delay VST with that features that support subsamples ?. I find that such thinks not called delay. thiey are called phase align and have a demo
https://www.forward-audio.com/phase-alignment-plugin-fatimealign/
for subsample delay need 3 interpolations which cause errors. but i test this plugin and hear no diffrence at 44 khz.
What i want say with the period time is more that human can mostly not hear 20 khz thats period 50 microseconds. but can hear 22 microseconds delay between left and right channel. also stereo hearing and reverb happen on mid range frequency. I read that above 1.6 khz the brain do this on diffrent level and not delay between left and right diffrence.
so it is logic for good stereo reproduce, that the mid range speaker is precise and not floppy so that the transients are produce at exact times. Question is how can see a precise mid range speaker in measurement witgh REW(when the step response not show this) ?.
bennybbbx
Addicted to Fun and Learning
ok, but rew seem not have cumulative spectral decay. i see decay values in REW but they seem not able to compare a fast and a slow speaker. headphone that reach low frequency is much faster as small speaker. the iloud mtm have more and lower bass as the Kali LP6 but the kali have a slower step response. the eris 3.5 have much less bass as the MTM but simular step response as the MTM. and it sound in stereo width not much diffrent to MTM. maybe the step tesponse change lots when use measure start frequency from 100 hz or form 2000 hz. but in real world speakers i see clear the small speakers are more precise. see my old post with the step resonse links to the speakers
Well, both good time (and phase) alignment and amplitude alignment between channels is the bigger problem, but indeed, removing non-delay parts of the midrange (and everything else) is a good place to start.
Remember that a phase shift that looks like phase=2*pi*f*t (where t is the time delay in seconds, f the frequency in Hz, and phase being the phase shift in radians) is exactly a time delay. Time delay and phase shift are inextricably linked. In fact, if you want to know the time delay of a process, you can do a linear fit on the unwrapped phase of the process, and there's your average delay. The slope at any given point provides the time delay at that given point, too (but it's a touch more complicated than just that).
Remember that a phase shift that looks like phase=2*pi*f*t (where t is the time delay in seconds, f the frequency in Hz, and phase being the phase shift in radians) is exactly a time delay. Time delay and phase shift are inextricably linked. In fact, if you want to know the time delay of a process, you can do a linear fit on the unwrapped phase of the process, and there's your average delay. The slope at any given point provides the time delay at that given point, too (but it's a touch more complicated than just that).
ernestcarl
Major Contributor
Well, there is... but how to set it up is not particularly straightforward it seems. You have to make adjustments to your filtering analysis and other view settings to get better insight with the plots. Application of some extra IR windowing/gating is also not straightforward as well as it seems one may need to export a filtered/windowed IR first then re-import it back into REW -- not sure if that is even necessary. I think CSD plots are really hard to compare between different sources because your own measurement parameters (how accurate are your measurements anyway?) as well as filtering view settings can be manipulated in so many ways.
bennybbbx
Addicted to Fun and Learning
I think you can get db values to set the cursor when you click on a frequency. in my screenshot i do this at 995 hz. the level is 88.7 db and 40 ms later it is 85.3 db. this is only 3 db after 40 ms. that look as nonsense. i also do measure with lower volume, the result was after 40 ms less db. so i thougt make loud can see more diffrence.
can you please upload your mdat file, so i can see what parameter you use ?. In your plot reach 60 db before 14.9 ms. this look more usefull. to compare csd then need a set of default setttings or need on all csd plots add the settings that is use
ernestcarl
Major Contributor
The cursor was pointed at the lower left corner so I would have ignored the dB level in green. Also REW may not be the best way to view CSD type of plots.
The measurement (10dB offset) was taken in not so ideal circumstances with the front port blocked and some EQ applied. Quite frankly, I do not think it is going to be of much use.
Yes, evaluating and assessing CSD is not easy. But, of course, it is easier than evaluating the decay behavior of a driver by looking at the step response.
The most important thing is the suppression of room resonances/reflections, because of course you don't want to evaluate the decay of the listening room. This only works via a gated window.
To be sure that the considered resonances in the CSD are really no room resonances, it helps if not only the possibility of a time-based representation is available, but also a oscillation period-based representation.
An example:
Midrange measured at 1m from cabinet without crossover.
1. CSD based on "optimal"** gated measurement
2. CSD based on bad gated measurement, a small part of a room resonance is included in the waterfall diagram.
3. same as 2 but as spectrogram
4. same as 2 but a CSD based on oscillation period in spectrogram representation
In the period-based representation 4, the spatial reflection can be identified as an angular decay process.
If you want to read more about this: I. Mateljan, H. Weber, A. Doric: Detection of Audible Resonances
In addition, the CSD is also strongly dependent on the sampling frequency of the measurement.
The whole thing is that simple
** as good as it gets
Last edited:
ernestcarl
Major Contributor
I think that’s what I thought as well with wavelet spectrogram view being somewhat easier to “read” and more useful (to me at least) in REW. But I’d rather leave the detailed loudspeaker analysis to others.
bennybbbx
Addicted to Fun and Learning
I have now find settings that show clear which is precise and which is not. i have add headphones too. because lp6 have crossover 1.5 khz i choose frequency 1 khz for all speakers to compare. the waterfal image is not very usefull. in decay i choose 1 ms slice interval and so the db values are display for 1 khz. so can clear see how much. I try lower values as 1 ms slice intervall but thats not possible. the kali have low volume. so you need always subtract the other values from the 1. value. soothing 1/3 i use because with smoothing 1/48 the frequency response change much and can not good compare. The eris 3.5 look best in this and i think they sound a little more clear as the mtm but i like the mtm more i think rooms sound more real. so i keep the eris 3.5 and the MTM and the canton 4 inch i sold. the canton sound clear not so good as the MTM and Eris 3.5 in stereo width.
Last edited:
tuga
Major Contributor
bennybbbx
Addicted to Fun and Learning
I think most important value is to use small rise time. they use 0,15 ms. your used risetime of 6 ms is much too slow to show speaker precision https://www.stereophile.com/content/kef-ls50-meta-loudspeaker-measurements
here you can see a slow 10 inch speaker and a fast electrostatic speaker in compare. I think that the magnepan speaker sound more stereo happen mostly because they are fast in the mid range ears need for binaural hearing. http://theaudioannex.com/forum/threads/how-to-read-a-speakers-cumulative-spectral-decay-plot.9439/
maybe you can measure your speakers with the settings they use ?. 0,1 ms rise time 4 ms time range and if you have good signal to noise ratio you need no CSD mode and so to reduce room influence you can set window size to 4 ms too.
and in REW manual stand this about CSD mode. i better not use it and use shorter window time. the curve on low frequency is because window time is too short for CSD mode.
Last edited:
ernestcarl
Major Contributor
Don't really know.
From Stereophile
30dB scale? Measurements: Fig. 2 Tannoy Revolution XT 6, cumulative spectral-decay plot calculated from output of accelerometer fastened to center of sidewall
JA's technique is quite different in that he uses an accelorometer.
(edit: forgot Amir had a Klippel CSD for the same coax speaker)
CLIO system from acoustic measurements look a little different (speaker uknown):
40dB scale. Rise time 0.58ms
Neumann shows theirs differently as well (KH120)
30 dB scale. No rise time specified and resolution probably is lower
Resolution Magazine do show CSD type looking waterfalls in their measurements
40dB scale. Presonus Sceptre S8
I could not quite see the ~550Hz resonance in my own earlier very nearfield measurements... but there was some around 470Hz or so instead, if I remember correctly.
I haven't really thought of it much as something to show speaker "precision", speed, or stereo accuracy as @bennybbbx. As a way to check for extended resonances or ringing, sure.
Last edited:
ernestcarl
Major Contributor
With the same mdat file measurement... It seems that 500+Hz resonance of the S8 can be seen clearer with those settings:
Though I'm not really sure that I should care much more about it than merely acknowledging that it is visible there in this view when zoomed-in really close.
tuga
Major Contributor
From Stereophile
30dB scale? Measurements: Fig. 2 Tannoy Revolution XT 6, cumulative spectral-decay plot calculated from output of accelerometer fastened to center of sidewall
JA's technique is quite different in that he uses an accelorometer.
He also uses a microphone for the speaker output CSD:
Fig.10 KEF LS50 Meta, cumulative spectral-decay plot on HF axis at 50" (0.15ms risetime).
Fig.9 Klipsch Klipschorn, cumulative spectral-decay plot on tweeter axis at 50" (0.15ms risetime).
Measuring Loudspeakers, Part Two Page 6
To produce a meaningful CSD, the time data need to be free both from noise—it helps to average as many separate impulse response measurements as the host PC can manage—and from environmental reflections, such as those from the microphone stand and its associated hardware, and the speaker support. If not, such reflections produce spurious ridges in the plot that might be interpreted as indicating the presence of resonances. Again, it is also important not to aggressively window the data and produce too short a time record. While this can produce smooth-looking plots [37], they are misleading. Version 10.0A of the MLSSA software flags the area in a CSD plot with dots where the data are invalid due to an inadequate time record (shown in the bottom-left corners of figs.18 & 19). Hawksford [38] has also suggested modifying the CSD plot by compensating for the loudspeaker's minimum-phase behavior. This should make lower-frequency resonances easier to see, but I have yet to try it.
https://www.stereophile.com/content/measuring-loudspeakers-part-two-page-6
Similar threads
- Poll
