Beyond the Ariel

does any of this matter?

What strikes me is whether or not any of these short duration artifacts (or any of the other 'directivity' measurements and concerns are actually audible w/o direct listening to designs with and without said variations under controlled conditions.

Lots of "technical" discussions going on in various threads, but not many listening impressions...

John L.
 
mige0 said:
What else is striking to me is that we see decay from mere interference (front and back wave) – which I thought (and have read about) isn't the case.


Calling the process "interference (front and back wave)" sounds strange to me. Isn´t it simply edge diffraction, which obviously would have to decay with the same rate as the original signal?

This would have the impact that every comb filtering effect is prone to decay as well and *not* only to FR deviations as it is exactly the same case here.

I would not see edge diffraction as a comb filtering effect between time-delayed waves.

Apart from that, thank you for a very helpful investigation, at least for me. :)

Rudolf
 
gedlee said:
A resistance ladder is a whole lot cheaper.

Excuse the ignorance, but do you mean an L pad, T pad, or something of that nature?




Lynn Olson said:
What I do like is that it buffers the wild impedance swings of the horn/driver combination, so the passive crossover can see a near-resistive load.

Amplifier -> Passive Lowpass Filter -> 9~10 ohm shunt to ground -> transformer primary -> transformer secondary -> optional Zobel inductance correction -> driver.

The transformer multiplies the impedance of the driver/horn by the square of the turns ratio, and the 9~10 ohm shunt to ground flattens out most of the reactance, leaving a load (as seen from the lowpass filter) that is close to resistive (8~9 ohms). From the perspective of the driver/horn, it sees a source impedance that is divided down by the square of the turns ratio - basically, a fraction of an ohm. Most of the reactance of the driver/horn is buffered from the passive crossover, which sees a load close to resistive.

Another stupid question, sorry: if the shunt resistor cancels most of the reactance, AND nullifies the driver resonances, what exactly is the advantage of the autotransformer? What does the amp seeing very high (I guess from your description above) impedance, and teh driver seeing very low impedance do that is desirable?
 
augerpro said:

Another stupid question, sorry: if the shunt resistor cancels most of the reactance, AND nullifies the driver resonances, what exactly is the advantage of the autotransformer? What does the amp seeing very high (I guess from your description above) impedance, and teh driver seeing very low impedance do that is desirable?

Let's walk through an example. Say we want 12 dB of attenuation. That's a voltage reduction of 4:1, so that means an autoformer or transformer will have a 4:1 turns ratio (transformer losses are typically 5%, so they can be neglected).

An example driver/horn combination might exhibit an impedance variation from 8 ohms to 40 ohms, and this variation consists of several peaks at and above the horn cutoff - falling in the crossover region of interest. These impedance variations will add a lot of complexity to a passive crossover network connected directly to the driver/horn system.

The 4:1 turns ratio multiplies the impedance 16 times, so the impedance appearing at the primary ranges from 128 to 640 ohms. This primary impedance is shunted by 8 ohms, so the net load impedance as seen from the passive crossover ranges from 7.53 to 7.9 ohms, a variation of 5%, too small to have a significant influence on the crossover.

What about the source impedance the driver sees? If the highpass filter is a 1st, 3rd, or 5th-order passive network, the impedance at infinite frequency is the source impedance of the amplifier + speaker cable, and the impedance at zero frequency is infinite. It is a complex reactance at all frequencies, as seen from the driver.

If we look at the lowest usable frequency of the transformer - say, 100 Hz - and assume a crossover source impedance of 80 ohms, the 80 ohms shunted by the 8 ohm-resistor presents 7.27 ohms to the primary, and the transformer converts that to 0.45 ohms at the secondary - this is the source impedance the driver sees. The source impedance always falls between 0 and 0.5 ohms, regardless of the topology or behavior of the crossover.

The combination of the shunt resistor and step-down transformer act as a buffer between the crossover and the horn/driver combination. It takes a liability of the horn/driver - more efficiency than you need, and a requirement to smooth out the reactive load - and makes it into an asset.

If there's a requirement for CD equalization, the transformer approach probably gets too complicated, since there's an overall tilt to the whole crossover/attenuator system. An L-pad with carefully chosen HF bypasses is probably the only way to go - or skip the whole thing and use active equalization.
 
Rudolf said:


Calling the process "interference (front and back wave)" sounds strange to me. Isn´t it simply edge diffraction, which obviously would have to decay with the same rate as the original signal?

I would not see edge diffraction as a comb filtering effect between time-delayed waves.

Rudolf


As I see it – any change in room available for a wave front propagating in space creates reflections of energy.
Smooth changes less – abrupt and large changes more – like a half way mirror.

Its this reflecting character I see as the root / cause / basic mechanism of diffraction of any variety – be it at the OB edge - or when entering a horn shape - or leave it into 4Pi - or at "non natural" shapes along its path (Earls attempt) – or inside the phase plug of a compression driver – or when a wave hits any border in your room (this is the most obvious case of (total) reflection though) where space available suddenly is less (not more)....

Its called honk (together with the amplifying mechanism of the horn) – or diffraction for OB – or reflection (of course) - or anything else – but to me its always energy reflection (as I had to learn over the course).

Where it happens is at the point where the wave front is bent – creating kind of second source – as if part of the waves energy is split into what's going on its way (slightly attenuated) and the rest acting as if radiated form that very point into all directions available – also back to the speaker or wherever.

The change of 2Pi to 4Pi certainly is a pretty sharp change in available space when the wave front reaches the OB border – creating strong second sources (forming a dipole by itself at the border of the baffle) as discussed a lot at the beginning of this thread.

What we hear at any position in our listening room is nothing else than an overlay of front wave + back wave + all diffractions happening.
It certainly is mere interference – no?

But where the heck is the "energy storage" mechanism we see in the CSD at 45 deg measurement?



soongsc said:
Since the front and back waves meet at different times around the edge of the baffle, the cancellation also occurs at continuously different times. It would be interesting of this could be modelled and investigated to see how driver location on the baffle effects the response.

Thinking about while writing – sh**t – its not an "energy storage" mechanism to look for from the OB itself – its the filter I created to get close to ideal and time accurate on axis response.
Yes - it perfectly cancels out the delayed arrival - but does that by creating an inverse delayed movement – which does not sum (interfere) to zero at any other place in space.



john k... said:


What you are seeing is typical of a dipole response when the frequency ranges extends up to and beyond the dipole peak. A dipole is only constant directivity starting about 1 octave below the dipole peak frequency. Above that frequency, but below the first dipole null, the polar pattern starts to broaden out (the figure 8 getts fatter) and there is an off axis peak around the frequency of the dipole peak. If you were to LP the response with corner frequency an octave below the peak (or even lower) you would see true CD response. provided the driver also radiated omni directionally (both from the front and rear side).


John, this would mean that even at frequencies where OB can be assumed to be constant directivity we would get added decay at off axis positions - but no change in FR.

How would that fit into min phase behaviour?



auplater said:
What strikes me is whether or not any of these short duration artifacts (or any of the other 'directivity' measurements and concerns are actually audible w/o direct listening to designs with and without said variations under controlled conditions.

Lots of "technical" discussions going on in various threads, but not many listening impressions...

John L.


Does decay or uniform power response within reasons count sonically? I think so.
But to tell the truth – even in the presented arrangement sound is great!

If we could manage to have a tweeter with the same directivity (fat figure eight) it even might be an advantage as we get more energy to the sides (like toeing in speakers extremely to get wider stereo image coverage)

I'd be very interested what would show up at comparable CSD measurements over different angles from horns.
Lynn are there some you already could show us?

Michael
 
Lynn Olson said:
An example driver/horn combination might exhibit an impedance variation from 8 ohms to 40 ohms, and this variation consists of several peaks at and above the horn cutoff - falling in the crossover region of interest. These impedance variations will add a lot of complexity to a passive crossover network connected directly to the driver/horn system.



On my waveguides there is only a single compression driver impedance peak at or above the crossover and with a "smallish" resistor shunting the driver the complex frequency response EQ and HP circuits see pretty much a flat impedance so they are highly predictable and effective.
 
mige0 said:


John, this would mean that even at frequencies where OB can be assumed to be constant directivity we would get added decay at off axis positions - but no change in FR.

How would that fit into min phase behaviour?


Michael

No. You get less delay off axis. For example at 90 degrees the delay between front and rear is zero.

The lower delay off axis has the effect of pusing the dipole peak high in frequency which results in the same 1st order roll off but with starting at a higher frequency which is why you see lower amplitude off axis in a dipole. The thing here is that all this is still MP relative to the observation point. There is a change in FR but if the LP filter is at a sufficiently low frequency this difference is above the LP cut off. You may see some small change in phase due to what is happening in the stop band but these will be less significant as the LP corner frequency becomes lower.

CD is CD and if the response is truly CD then on and off axis the phase will be the same.
 
Thanks John - sure - its gradual - I was too quick with assumptions...

Earl, do you have CSD plots of the OS waveguide you could share?
Telling from FR they should look pretty much the same over different angles - but I find CSD most intuitive to read.

One thing I'd like to add - telling from my measurements above - a 8" mid isn't too big with respect to beaming as the baffle width still is the dominat part here - and if such a speaker behaves well to the upper XO point - as is the case here, its a design choice really worth to consider.

Michael
 
mige0 said:

Earl, do you have CSD plots of the OS waveguide you could share?

Michael

I don't take CSD data, because I am not yet convinced that anything is visible there that is not visible in the impulse response itself. I have supplied impulse responses to others before and they did CSDs on them, but alas, because I don't do them, the noise floor in the data wasn't very good and that made the CSD look bad. Some attacked this as a problem in the waveguide, which it was not.

As to CSD, I am changing my point of view - a little. The impulse response DOES tell all, but being linear in amplitude, it is not good at showing audible effects that occur later in time and hence much lower in amplitude. What is required is a more log amplitude display as this is more in line with what the ear actually does. CSD does provide this log magnitude function and thats positive, but I'm not sure that its necessary. For example, one could just plot the impulse response with log(|p|) and that might work fine as well. But any of these techniques will be highly susceptable to the noise floor and most of my data is not that great in this regard.

Thats said, I may be able to find some to share if you wanted to run the CSD yourself.
 
Regarding signal to noise ratio for CSD measurements.

The measurements I have shown recently have been taken in doors - but with no special care to ambient noise.

I may have had roughly 40 - 45dB noise floor which leaves 45dB S/N for the measurements at 90dB SPL.

This is not great at all - but still comfortable for the intent. We don't have to be too picky in this IMO.


Furthermore you immediately see the noise floor if we change unsual 25dB scale of CSD to - say 50dB.
Its really a *floor* that is displayed, if you have a PC rumming close by also the dominat frequencies of the vent and HD as something that sticks out and does not decay.

So very easy to identify...


Michael
 
Ok - I remember you having once said that you do not use commercial software (like ARTA for example) for your analysis but do the coding by yourself (in mathcad IIRC).

I guess the crosscorrelation you mention is related to that process?
Does it mean that the artefacts of calculation done there are not evenly spread like a "floor" - making CSD post processing from your impulse extraction less usefull as usual?

Not sure if I got your point.

Michael
 
Courtesy Earl

OS wave guide at 0 deg:

An externally hosted image should be here but it was not working when we last tested it.



OS wave guide at 15 deg:


An externally hosted image should be here but it was not working when we last tested it.



OS wave guide at 30 deg:


An externally hosted image should be here but it was not working when we last tested it.



OS wave guide at 45 deg:


An externally hosted image should be here but it was not working when we last tested it.




48kHz Files provided by Earl I converted in CoolEdit and processed by ARTA to get CSD.
Looks pretty much excellent for me with respect to decay time
:)

Michael
 
Disabled Account
Joined 2009
gedlee said:
They will generally appear as a floor, its not caused by room noise thats all. My room noise is quite high with no end of noise interuptions from cars, airplanes, kids, HVAC, you name it. Without noise imune signal processing the data would be quite useless.


:confused:

Noise will have its impact regardless of cross correlation. What then is "room noise" opposed to what? Not caused by room noise but shows as a floor. Room noise is ... example given: airplanes. A bit confusing to me.

CSD is quite useless. One would have to know which deviation from ideal is critical or not. The threshold of perception, related to the graphs is where? CSD is nothing but the impulse response. To estimate the impact on audio reproduction the better way is to look at (1) amplitude over frequency and (2) group delay. CSD is nothing else, really. It's apperance leaves room for some free style interpretations. But that means nothing but confusion. Every technical info on audio transducers should be related to psychoacoustics. But it can't be done - honestly - with CSD pictures. CSDs can be compared to each other. But what does a difference tell about sound quality?

have fun
 
Hello mige0,

It seems that the window rise time of the CSD you obtained using Arta on Earl's pulses is not the same that you used previously for the CDS in your message:

http://www.diyaudio.com/forums/showthread.php?postid=1818279#post1818279

Could you indicate the parameters used to obtain those CSD:

FFT size
Window rise time
Slice shift
Smoothing

(you said in your CSD comments that the decay was excellent, could you specify the frequency above which you consider the decay being excellent...)

Best regards from Paris, France

Jean-Michel Le Cléac'h


mige0 said:

Looks pretty much excellent for me with respect to decay time
:)

Michael