[Kevin Haskins]

I've had comments about measurements.   Usually the questions center around someone looking at a measured response somewhere on the web and comparing it to one of my measurements.   Its tough to do A/B comparisons based upon different measurement set-ups.   Anyone who does this a lot knows there are plenty of ways of cooking the data.

Here is just a simple example.   Here is on on-axis measurement of the Kepler.




Here is a frequency response measurement that looks much worse.  Your first inclination is to say that this measurement is pretty ugly.   It would be a bad inclination though because this is the same measurement.   This one is just 5db per division and we are zoomed in to look at just a few divisions.   This makes the response look much uglier.   For this reason, in most marketing materials you will see 10db per division measurements.    I tend to show mine in 5db per division and 12 divisions on the screen.  Why?  Well.... because Praxis defaults to that mode and I'm too lazy to change it.   



Just a public service announcement... we will now go back to our regularly scheduled program.

[JohnR]

And don't even get started on smoothing!

[Kevin Haskins]

And don't even get started on smoothing!

I'd say 1/3rd octave is legit.   If you want to really strut your stuff 1/6th octave is fine.   For crossover work its terribly difficult to do much with small aberrations in the frequency response so anything with more resolution than 1/6th octave is of limited use.   Our ears have some smoothing built into them so using it on measurements is legit because its reflective of what we actually hear.   

Erroneous data is just as bad as wrong data.

[Danny Richie]

I'm pretty used to those 5db scales myself and that response looks pretty good to me.

[Kevin Haskins]

I'm pretty used to those 5db scales myself and that response looks pretty good to me.

Thanks Danny.... you cannot do much about diffraction ripples.   Move off axis 15 degrees and they mostly go away.

[TerryO]

I'm pretty used to those 5db scales myself and that response looks pretty good to me.

Thanks Danny.... you cannot do much about diffraction ripples.   Move off axis 15 degrees and they mostly go away.

Mr. Haskins,

I've found, through extensive research, that by moving the on-off switch approximately 45 degrees, you can make them completely disappear.
A lot of people, who are not Audio Professionals, lack the experience and knowledge to understand these esoteric details.

Best Regards,
TerryO

[Kevin Haskins]

I'm pretty used to those 5db scales myself and that response looks pretty good to me.

Thanks Danny.... you cannot do much about diffraction ripples.   Move off axis 15 degrees and they mostly go away.

Mr. Haskins,

I've found, through extensive research, that by moving the on-off switch approximately 45 degrees, you can make them completely disappear.
A lot of people, who are not Audio Professionals, lack the experience and knowledge to understand these esoteric details.

Best Regards,
TerryO

Some speakers sound a lot better with the switch "off".   

[jose]

This somewhat off-topic, so I apologize, but I have always wanted to ask the experts: why are anechoic measurements taken at 1 meter? Doesn't it make more sense to measure at a typical listening distance, which is normally about three times that?

Thank you,
Jose

[Russell Dawkins]

My favorite recent example of misleading graphing is 2/3 way down this page:

http://tinyurl.com/2k8cuy

It's the in room response of the Feastrex/Augie OB by Dick Olsher.

Looks reasonable until you notice the scale. 20dB/division! 

[HAL]

Looks like 10dB/Division to me. 

[gitarretyp]

I think he means the second plot, which is of the full system.

[Russell Dawkins]

I'm not referring to the first chart 1/4 way down the page (the Augie in the Visaton baffle) which is the standard 10 dB per division, but the in-room frequency response, second to last, 2/3 way down.

Look on the left side of the graph; 35 dB, 55 dB, 75 dB, 95 dB, 115 dB bottom to top, going one division at a time.

[HAL]

See it now.

Thanks.

[Russell Dawkins]

Just imagine what that graph would look like at 5 dB/division!

[Kevin Haskins]

Just imagine what that graph would look like at 5 dB/division!

Yes.... it is about +/- 5db from 100Hz up in-room.    You would need two of those 5db divisions to even look at all of it.   

[DevilDriver]

A fine example of how misinterpretation of data can lead to false conclusions.  One of many possible problems.

I thoroughly believe that every manufacturer should be straight forward in providing as many measurements and objective data as possible.  But this obviously requires that each customer understand how these results are derived and what the results really mean.  Unfortunately, this requires a ton of customer education that is simply impractical...sometimes I think we're better off keeping the measurements at an industry level because a single customer who doesn't fully understand what they're reading can run to 100 different forums and talk about how such and such driver measures poorly and should be avoided.

Ok, that was a bit of a rant, but I think the point is clear.

[Kevin Haskins]

I agree... there are some things I don't post for that reason.  There is such a thing as too much information.   I don't post Klippel measurements because you really need to know what your doing to interpret the data.   Most of my competitors don't know how to read them, how is a consumer going to get meaningful information from them?

Distortion measurements are another one.   Customers don't understand them and for transducers, they are relatively high (maybe x100) compared to what people see from amps and other components.   

Another factor is that you need a chamber to do valid distortion measurements.  I'm borrowing one next week for some measurements of my speakers but its not something that I typically have available for use, nor do 99% of the speaker designers.   So... the data by itself is meaningless.   It has to have context before consumers can use it.

[Bob Reynolds]

And don't even get started on smoothing!

I'd say 1/3rd octave is legit.   If you want to really strut your stuff 1/6th octave is fine.   For crossover work its terribly difficult to do much with small aberrations in the frequency response so anything with more resolution than 1/6th octave is of limited use.   Our ears have some smoothing built into them so using it on measurements is legit because its reflective of what we actually hear.   

Erroneous data is just as bad as wrong data.

What exactly is 1/3 octave smoothing? What's the computation?

Dr. Toole states in his white papers that in the bass region (below 200 Hz) 1/10 octave resolution is necessary to really see what is happening in-room so that EQ can be properly applied.

[Bob Reynolds]

I agree... there are some things I don't post for that reason.  There is such a thing as too much information.   I don't post Klippel measurements because you really need to know what your doing to interpret the data.   Most of my competitors don't know how to read them, how is a consumer going to get meaningful information from them?

Distortion measurements are another one.   Customers don't understand them and for transducers, they are relatively high (maybe x100) compared to what people see from amps and other components.   

Another factor is that you need a chamber to do valid distortion measurements.  I'm borrowing one next week for some measurements of my speakers but its not something that I typically have available for use, nor do 99% of the speaker designers.   So... the data by itself is meaningless.   It has to have context before consumers can use it.

I really appreciate the effort knowledgeable people put forth to educate us consumers. So I don't agree that there is such a thing as too much information. I may not understand it all upon the first read (or after 100 reads), but making it available is like a good reference text - I can refer to it as I need.

Distortion measurements are something that I wish all speaker manufacturers published (and that JA at Stereophile measured). As you point out the values for drivers (woofers) are quite high relative to typical electronics. I think as more people understood this, we could get away from sweating the details of electronics and focus on speakers. Plus, I think it would push more audiophiles to embrace subwoofers. Would you give us an explanation as to why an anechoic chamber is necessary for accurate distortion measurements?

Many thanks.

[DanWiggins]

I'm not Kevin, but maybe I can shed some light...

What exactly is 1/3 octave smoothing? What's the computation?

Typically the sum of all linear points between the limits of the octave, divided by the number of linear points within those limits.  Assuming your limits are 100 and 125 Hz, and you have 15 points, you would add those points together, divide by 15, and assign that to the octave.

For high resolution linear FFT systems like Praxis (and most other high-end measurement gear), you center an 1/N octave bandwidth on a given point, and run your filter.  Repeat this for all points in your dataset so that every linear frequency point now has a corresponding 1/N octave smoothed point.

Dr. Toole states in his white papers that in the bass region (below 200 Hz) 1/10 octave resolution is necessary to really see what is happening in-room so that EQ can be properly applied.

For high resolution post-crossover EQ systems (like the Woofer Widget from Kevin), you want to run very high resolution sweeps; I like to use sub-1 Hz steps (say a 32K FFT on 24 kHz sampling rate data). 

However, when doing a passive crossover for a loudspeaker system, trying to affect response anomalies finer than 1/6th octave is nearly impossible; you simply cannot get the high Q results required from standard caps, coils, and resistors.  So looking for 1/3rd octave issues and addressing them when voicing the crossover is typically the best you can hope for.

Would you give us an explanation as to why an anechoic chamber is necessary for accurate distortion measurements?

Two reasons: resonances and noise.  Resonance can artificially increase the amplitude of either a fundamental (thus lowering the true THD) or select harmonics (thus artificially increasing THD).

Most THD is typically measured by applying a brickwall notch filter to the excitation tone, and all remaining audible content is considered THD plus Noise (THD+N); usually this is referred to as simply THD, since for passive systems the addition of noise is essentially impossible (speakers don't really have noise).  Spurious sounds - like 120 Hz flicker from a lightbulb, or the hum from a computer fan - will be added in the summation, leading to a higher level than actually generated by the speaker.

An anechoic chamber solves both problems.  An anechoic chamber is, by definition, free of resonances/echoes down to a given frequency.  The chamber Kevin will use next week is anechoic down to 35 Hz, meaning that THD measurements above that point are very clean.

An anechoic chamber is also very well isolated from external noise sources; the chamber being used next week has a 9 dBA background level (meaning that the average background noise level in the chamber is less than 400 uPa of noise pressure).  Consider that a VERY quiet recording studio is around 15 dBA, and a quiet office or conference room is around 45 dBA.  So when measuring THD or other measurements at levels around 70+ SPL, the impact of noise is greatly reduced.