Step Response: Does It Really Matter?

[FeddyLost]

Can step response be used to evaluate detail retrieval from a speaker, comparing the difference of the anechoic response compared to a standardised "normal-room" setup?

It's very troublesome to predict resulting psychoacoustical "detail retrieval" into brain.
I have very (excessively) dry room, and I'd say it's not the best solution for detail retrieval in typical complex constant sound field like typical music. Proper diffusion must work better according to all opinions at GS.
It will also alter step responce somehow, but explanation will require a lot of research with test subjects and complex stimulus.

 

[Andysu]

I have not read the thread in detail, but I have a general question. Can step response be used to evaluate detail retrieval from a speaker, comparing the difference of the anechoic response compared to a standardised "normal-room" setup? I was wondering if the difference signal could be used to extract early reflections within 2 ms and be used as a "detail index"?

Below step response 50 cm from a speaker placed near a wall.
View attachment 126420

Below step response measured at listening position.
View attachment 126421

You say mic near a wall. Show the pictures of your room and where you placed the mic, so repeat the tests and take pictures of yourself in the room where the mic how you did this, thanks.

 

[Thomas_A]

You say mic near a wall. Show the pictures of your room and where you placed the mic, so repeat the tests and take pictures of yourself in the room where the mic how you did this, thanks.

No, not mic near wall: speaker against the wall, with a damping panel behind. Mic 50 cm from speaker vs. at LP.

 

[Thomas_A]

It's very troublesome to predict resulting psychoacoustical "detail retrieval" into brain.
I have very (excessively) dry room, and I'd say it's not the best solution for detail retrieval in typical complex constant sound field like typical music. Proper diffusion must work better according to all opinions at GS.
It will also alter step responce somehow, but explanation will require a lot of research with test subjects and complex stimulus.

Diffusion would be more related to envelope and stereo system errors (filling the comb filtering) and also later than 1-2 ms. I suspect. These are important but for other reasons.

 

[bennybbbx]

important information about REW impulse export. when export the impulse there is no microphone cal correction do. here can read, hope come in future soon. https://www.avnirvana.com/threads/i...e-as-wav-do-not-work-correct.8735/#post-67118

so my impulse export wav i upload here are not so good.

 

[Tom Danley]

Like the familiar magnitude and phase, impulse response and step response are different views of the same event. I don't find that feature very informative and unless the measurement is "long enough" it would be wrong to associate it with anything the woofer is doing.

What is useful and sometimes helps locate an issue in the speaker or reflections is what Dick Heyser called the "energy time curve" and now more often called "Envelope time curve". This is more or less loudness vs time" I attached a screen shot of an ARTA measurement i had handy which shows the impulse response and the ETC. One can see the floor bounces and stuff at the couch much more clearly than the impulse shows.
Best;
Tom

 

[FrantzM]

Like the familiar magnitude and phase, impulse response and step response are different views of the same event. I don't find that feature very informative and unless the measurement is "long enough" it would be wrong to associate it with anything the woofer is doing.

What is useful and sometimes helps locate an issue in the speaker or reflections is what Dick Heyser called the "energy time curve" and now more often called "Envelope time curve". This is more or less loudness vs time" I attached a screen shot of an ARTA measurement i had handy which shows the impulse response and the ETC. One can see the floor bounces and stuff at the couch much more clearly than the impulse shows.
Best;
Tom

Welcome!
Pleased to see you here Tom. Post more!

 

[bennybbbx]

Like the familiar magnitude and phase, impulse response and step response are different views of the same event. I don't find that feature very informative and unless the measurement is "long enough" it would be wrong to associate it with anything the woofer is doing.

What is useful and sometimes helps locate an issue in the speaker or reflections is what Dick Heyser called the "energy time curve" and now more often called "Envelope time curve". This is more or less loudness vs time" I attached a screen shot of an ARTA measurement i had handy which shows the impulse response and the ETC. One can see the floor bounces and stuff at the couch much more clearly than the impulse shows.
Best;
Tom

the envelope time curve in arte seem simular to the spectrogram with slice view can use in REW. REW is in this way enhanced and can show this for a choosable frequency.

question if the klippel have a simular display feature ?

 

[Tom Danley]

Hi
One big issue that makes this harder to grasp in a broad band sense is that for a given event, at 20Hz it takes 1000 times longer than the same event at 20Khz.
This for example makes the "impulse response" weighted and essentially blind to low frequencies or a better way to say it is that the LF end takes many milliseconds while at the top end, it's MUCH shorter. It is the difference between equal energy per Hz and equal energy per octave.

This also makes a CSD / cumulative spectral decay less informative than a "Burst decay" or wavelet view or STransform, these compensate for the time / wavelength difference vs frequency. With Burst decay a given event at 100Hz, has the same length in the display as the same event at 10Khz.

Boy i really like this forum! I used to post a lot on DIY forums but didn't like the arguments and the encroachment of audio VooDoo in service of sales. My interest is in how loudspeakers actually work and that's the path to good designs. That's what delightful about this place, it's most definatly not about magic knobs, cable towers and the other things that require knowledge / belief to "hear" but are undetectable without.
Bravo Amirm and team.

 

[Music1969]

Hi
One big issue that makes this harder to grasp in a broad band sense is that for a given event, at 20Hz it takes 1000 times longer than the same event at 20Khz.
This for example makes the "impulse response" weighted and essentially blind to low frequencies or a better way to say it is that the LF end takes many milliseconds while at the top end, it's MUCH shorter. It is the difference between equal energy per Hz and equal energy per octave.

Hi Tom,

Is this (part quoted) why step response is better than impulse?





Taken from @mitchco's great article:

https://audiophilestyle.com/ca/bits-and-bytes/what-is-accurate-sound-r923/

 

[KSTR]

While step response (== impulse response convolved with the unit step function) helps LF visibility in that it has an infinite" bass boost" (infinite 6dB/oct rising slope in terms of spectrum), it doesn't solve the time resolution problem of a linear time scale. A cycle of 20Hz is 1000x longer than one cycle of 20kHz, just like Tom said. So, a plot would need to have at least 4000pixel resolution to even see the sampling points near 20kHz... which poses the next problem. The plot must use sinc interpolation between sample points, and to have high resolution of the real event at HF one better use a 100Khz mic and a 300kHz++ sampling rate.
Many years ago I've seen attempts to use a log time scale for IR/SR but looks like the concept didn't stick.

 

[Tom Danley]

Hi
Ok, I think I can shed more light on this.
Basically we are normally dealing with a loudspeakers magnitude response, logical, as this is the primary thing but not only things we hear.


With a real crossover, one has to deal with and include the upper and lower drivers frequency response and phase, to make it simpler, we will ONLY consider the crossover filter portion here.


First, simulate Speaker1, a perfect single driver loudspeaker, lets say who's bw is that of a sealed box, response flat to 30Hz and has a 4th order low pass at 20Khz.

I did this here with a separate high pass and low pass block, summed together and fed into a perfect Naryabump EL Flatatron transducer.


The loudspeakers acoustic phase is shown with all the fixed delay like time of flight to the microphone removed. This is the phase response that indicates where the driver is in time relative to another frequency in time.. You see it follows the minimum phase relationship between mag and phase that nearly all single drivers mostly obey over some bandwidth.


Speaker2 is a two way speaker made with two more perfect drivers with the same bandwidth and also has a 2nd order high pass and low pass L&R crossover at 500Hz.
The following displays are all different views of the same events (like how the impulse response and mag and phase are tied through the FFT and IFFT process), the one not normally shown with loudspeakers is the square wave response.

This is a complex wide bandwidth signal where it is also easy to recognize corruption or alteration visually.


For a single source like in an electronic signal, one can read the square wave something like this.

An angled top and bottom reflects insufficient LF response, flat mag and phase response to DC would have a flat level top and bottom.

The vertical portions are tied to the HF bandwidth as the upper bandwidth is limited, the square eventually becomes a triangle wave when fully "slew limited".

To look "perfect" on an oscilloscope, one needs flat mag and zero acoustic phase from about 1/10 the F to 10X the square wave, frequency making it a hard signal to reproduce well or over a broad bandwidth.



So what if we have a normal 2 way loudspeaker with two perfect drivers with exactly the same frequency response, is that different?


Well with all the named filters one can get flat magnitude response and what one sees is that there is a phase shift from high to low that ultimately reaches 90 degrees times the filter order. Here it is a 2 way speaker 2nd order L&R HP / LP crossovers at 500Hz.


I matched the frequency responses so one can see that in addition to the normal minimum phase relationship (where any change in frequency response has a corresponding phase response) one also has an extra 180 degree "all pass" phase shift (phase shift without a change in amplitude) that the crossover has added.
You can see that the impulse response has changed a little bit but the step response is pretty different. I hesitated to put Group Delay in, too many people look at the numbers and don't understand them and think subwoofer are bad because big numbers seem like bad juju.... Group Delay means what it sounds like, the difference in time between one group of frequencies vs another.



It displays a delay and cannot go negative (requiring the illusive predictive circuit for that).

In Group delay, one can see a different view of the all pass phase shift where the lower portion right below crossover is later in time than the upper part.

The thing one must keep in mind though is that what 1ms means depends on / scales with the frequency. In other words, any given filter slope or change in response shape at 10KHz produces 100 times more group delay at 100Hz, anything at 20Hz lasts 1000 times longer than 20,000Hz


With the Step function and especially the Square wave signal, one can see the transmogrification of the wave shape where some of the frequency components are in a different "time zone" (shifted in phase) compared to others.
When using FIR DSP to correct loudspeaker phase, what one does (I like Re-Phase) is delay everything back to the farthest behind (normally lowest frequency) part you have Taps and allowable latency to correct to. In this speaker, one could FIR to correct the phase to equal speaker 1 and so you have the same group delay, square wave responses etc.
I hope i can post all the curves, if on they will be in a second post. Ok they are in a different order, look at mag, impulse step, gd square

Best,

Tom Danley

 

[bennybbbx]

I have now measure the JBL 104 BTW. this have the fastest raise and release time of step responses i measured and it sound best in stereo wide. it is a 4.5 inch coaxial system with crossover 1.7 khz. I measure the step response with , without EQ and at 1.5 cm distance(from grill). the fast result stay same. intresting is this speaker have no overshot in tweeter and i think the tweeter sound not harsh as other and have best transients and clarity. the delay of woofer is simular to the kali LP. but the LP6 sound very worse in stereo width. so seem the delay is not so important. fast decay time and no overshot make a good speaker i think. THe Eq i use in the JBL do mostly boost bass. Peak 50.8 hz, Q1.19 12 db. the 1.5 cm measure is wrong text. it have eq for 50 hz but no other frequency. as you can see in the JBL impulse i use invert impulse. the JBL do invert the signal. so keep in mind when use a subwoofer

and here is the LP6 . it have much overshoot and large release time of impulse

 

[audio2design]

This all intuitively makes sense from a pure physics point of view.

But it doesn't seem to map to subjective experiences. There doesn't seem to be a correlation between superior step response and blatantly superior subjective evaluations.

Not sure I have the fortitude to go through 7 pages ( make that 15) on this topic but should be evident that outside the ITD frequencies (200-1500Hz or so) the ear/brain does not appear to have time related mechanisms. Timing needs to be consistent channel to channel at given frequencies to avoid location based filtering irregularities, but otherwise outside the ITD frequencies it is not evident from a neural acoustic standpoint that phase, i.e. step response is critical.

 

[bennybbbx]

Not sure I have the fortitude to go through 7 pages ( make that 15) on this topic but should be evident that outside the ITD frequencies (200-1500Hz or so) the ear/brain does not appear to have time related mechanisms. Timing needs to be consistent channel to channel at given frequencies to avoid location based filtering irregularities, but otherwise outside the ITD frequencies it is not evident from a neural acoustic standpoint that phase, i.e. step response is critical.

when the ITD frequency go only upto 1500 hz, then the phase still need exact in microseconds. because left and right waveform of music is much diffrent so it can happen that the audio signal of the left channel for next samples need only change by value 2 on other channel by value 200. a delay of 0.1 ms is 10 khz and for ITD 10 degree. and mostly on 2 way speakers the bass woofers only can do 6 khz and then level go much down. then it is clear that 2 speakers that work with the independent left right signal never can reach the peak of the wav at correct time 0.1 ms time diffrence between left and right speaker. at the 800 hz reverb signal for example. you maybe think such high freq are handle by tweeter but it do not. can easy test when use 800 hz and reverb. then reverb have only 800 hz. but the brain need the exact values to create the correct 3d room size.

also nobody knows how ear and brain excactly work. interaural level diffrence can better hear when a signal is loud. there is often told(on big 2 way system( a speaker sound only good and free when make loud
. this maybe are the slow speakers. because they are too unprecise the brain cannot do ITD but when they are loud the brain can hear larger level diffence between left and right ear. so the brain can maybe create the room sound image better with levels. every record have a reverb add. I notice that with my best speakers (the JBL and the eris 3.5 as fullrange without tweeter) the room feeling is very good even when hear not loud. the lp6 get better when hear louder. on low volumes it sound as if there is no room and is record as mono
it seem clear that a speaker that have slow rise time and decay time cannot produce it good. for room size feeling frequency 200-800 hz are very important. reduce this frequency with eq and you hear(when you have a fast speaker) or headphone. on slow speakers hear this not so much. slow speakers work as a reverb canceler.

 

[Thomas_A]

I wonder if the tilme from tweeter to mid peak intertime is varying due to crossover freqeuncy of type of filter, eg secod, third or fourth order? I’ve seen varying response between ~250-500 us p-p.

 

[j_j]

Impulse response is better conditioned, but capturing it requires more sophistication, and not just using a single impulse. Back on the www.aes.org/sections/pnw web site somewhere there is a talk on "What is an FFT", it has code for you to capture your very own impulses responses in a reasonably clean fashion.

 

[audio2design]

when the ITD frequency go only upto 1500 hz, then the phase still need exact in microseconds.

Do you have any proof of that or even a good hypothetical reasoning? I didn't find one in your post to be honest. ITD is time differential of the same signal between two ears. That only requires consistency between the channels not absolute accuracy. Hence that does not even justify the requirement for perfect impulse response though I will accept that for channel consistency taking care of ITD frequencies from a time standpoint may be required.

We don't know all the inner workings of the auditory system, but we know a lot including what frequencies appear to be used to extract what information including timing. Given the lack of correlation between impulse response and subjective impressions we probably should be working to understand why that is as opposed to assuming it is wrong or at least wholly wrong.

 

[audio2design]

also nobody knows how ear and brain excactly work. interaural level diffrence can better hear when a signal is loud. there is often told(on big 2 way system( a speaker sound only good and free when make loud
. this maybe are the slow speakers. because they are too unprecise the brain cannot do ITD but when they are loud the brain can hear larger level diffence between left and right ear

Most recorded music is rather devoid of ITD information with even live recording sadly lacking. The most obvious ITD in recordings is often artificially generated.

Typically the difference between quiet and louder is simply varied response ala Fletcher Munson curves.

 

[j_j]

You are confusing several issues here. Intraaural phase shift only matters if it's fast enough to disturb the local waveform on the cochlea. In general, 15 degrees per ERB is ok.

IntERaural delays can be detected down to 10 microseconds. There are arguments that this only applies up to some lower frequency, which is only true if the envelope of the signal is flat, which is to say a 10kHz sine wave presented in headphones has no interaural time worry.

HOWEVER when you put an envelope on that signal, i.e. amplitude modulate it, now you can hear interaural time delays.

As to ITD's in recordings, there OUGHT to be ITD's in recordings, but 60+ years of practice and advice related to profoundly obsolete technology have pretty much prevented that for the time being. It's something that needs to be considered, but the resistance and outright scoffing from various people who want to protect their obsolete methods has pretty much prevented it to date.