Yes it is true. Audio Science Review will start measuring speakers soon. This category of products is by far the most requested when it comes to future work. Alas, measuring speakers is not the same as electronics. The high-end standard is to use a multimillion dollar anechoic chamber so that the results are room independent. Space can be rented from such a facility but it all adds up and is tedious to shuttle speakers back and forth to the chamber.
Alternatives have been around forever that both reviewers and DIY speaker builders use. These can approximate the speaker performance. Doing so however is quite a lot of work, often requiring measuring speakers outdoors and on tall polls and such. If one were to produce a handful of measurements a year, and could find someone who will do such manual work for little money, it could be done. I am neither. There are thousands of models and brands of speakers. No way can we remotely characterize a useful subset of speakers someone may want to buy using these methods.
Importantly, there has been excellent research correlating a weighted sum of various speaker measurement angles to listener preference. This work has been standardized in the form of CEA 2034 "spinorama." This research though requires access to anechoic chamber measurement and a hemispheric set of measurements all around the speaker. At Harman this is done with a microphone array and rotating the speaker underneath it. Here is a picture I took while I was there:
A turntable under the speaker rotates it and snapshots are taken where the microphones are. Once finished, you can approximate the sound distribution all around the speaker. Then, by researching what percentage of sound arrives at a listener from direct sound (in front of the axis of the speaker) and reflections, you can come up with a graph that can predict how a speaker can sound in a typical domestic room. As implied, direct sound which is usually the only measurement published makes up a portion of what you hear. You hear tons of reflections from the room and the sum total of these determines the tonality of the speaker.
Due to complexity and cost of the above, only a handful of companies have been deploying these measurements and almost none provide that data to public (they are used in design stage of the speaker).
A few years ago the German company Klippel, know for their excellent analysis tools and systems for drivers tackled this problem using a different system (called "Near Field Scanner" or NFS). Instead of moving a speaker and using an array of microphones, it keeps the speaker stationary and uses a "robotic" system to move the microphone around. It uses to very unique and highly useful methods:
1. It makes measurements in near field meaning right next to the speaker. This is sharply increases the signal to noise ratio of the measurements since sound power drops rapidly with distance. Normally doing so would give you an incorrect measurement of what you will hear when sitting far from the speaker. Special signal processing is used to project far-field (where you may be sitting) from near-field measurements.
2. Using advances math, and making dual set of measurements, it can separate the direct sound from reflected one. In doing so, it gives you anechoic measurement in any ordinary space!
Before explaining more, here is a quick video of the system from construction to use:
Since the system is fully automated after some manual setup, you can test a typical 2-way speaker in about 50 minutes. Once there, you get a wealth of information about the speaker. In the standard setup, it defaults to over 500 points which each measurement taking less than 6 seconds. Here is an example run:
And the same in 3-D space:
Here is a shot showing the impact of room reflections on measurements:
The red curve is what you would normally measure. The blue is the corrected one where the reflections (shown in dashed lines) is removed. This is a little monitor speaker though so no response in bass. If there were more, it would likely pollute the measurements more.
The signal processing error increases with frequency (which can be offset partially by increasing the Hankel function order). Fortunately there is no need to measure the full 20 kHz that way. Using the same technique DIY methods use (i.e. gating the measurements so reflections are discarded) gives us the equivalent anechnic data. That is shown with the vertical dashed orange line.
Back to CEA 2034 Spinorama data, Klippel NFS automatically generates them for you:
You want to have a flat on-axis response ideally (black) and smooth off-axis response. If you did that, then your "early reflections DI" would be a smooth line. It is not here:
Note that I have reversed the colors in the bottom chart. You can ignore the sound power directivity as that has no correlation with listening preference (it includes all reflections and direct sound whether you are likely to hear their contributions or not).
We get cool drill down such as these:
So if your room is not a rectangular box, you can decide if you should or should not care about some reflections.
You can also slice up the reflections vertically and horizontally:
Sometimes direct axis will not be good but if you go up or down (your ear relative to speaker tweeter), you may get more flat response.
You also get a bunch of pretty 3-D graphs like these 3-D balloons at any frequency, direction and distance:
Or plot all the frequencies at once:
You can see that as the frequencies increase, the sound becomes more directional in irregular ways as the directivity index showed us earlier.
Compare this to the data manufacturer provides (which is above average):
And yes, classical measurements like impedance versus frequency and phase, distortion, etc. are all available.
Amir Buys the Klippel Near Field Scanner (NFS)
Lots of goodies, right? Well, it was not all good news. The hardware and software with all the options retails for US $90,000. Add a computer, amplifier, miscellaneous items and you are up in $100,000 range. I started a discussion with Klippel and their very helpful distributor in US, Warkwyn a few months ago. After much discussion and great support from the company include founder Wolfgang Klippel, we arrived at a financial arrangement that would not cause someone declaring me totally insane for purchasing this system!
The package which weighed around 300 pounds arrived by air and cleared customs after Christmas. Unlike the video above, I assembled 99% of the unit myself with just an occasional help from my wife for heavy items. Did not want to impose on her any more than I had already done with this expensive purchase! The assembly took about 3 to 4 days, with 12 to 15 hour days. You may have noticed me doing electronic reviews near midnight for this reason.
In order to be able to measure floor standing consumer speakers, I had to purchase the "high Z" version which means the measurement rig can climb to some 12 feet! And needs a large radius around it to boot. This meant I only had one place to put it which was our garage. With winter here and temps 10 to 15 degrees above freezing. and concrete floor, it was not fun. But I managed through it.
Next was getting the software and hardware to work together to create my first measurement. With everyone being on holiday, I had to resolve most of the problems myself although Klippel support in Germany was excellent.
Typical of pro software made for just a few people to run, it doesn't run by just clicking and hitting go. I had to get used to the interface of Klippel software which is ongoing as I type this. I have now run a dozen measurements to a point where I feel kind of comfortable with the system end to end. There are a lot of degrees of freedom from the robotic system setup to initial configuration and way the software runs to get usable measurements.
Road Ahead
No, I am not ready to measure any speakers yet. I have to get more comfortable with the data being generated by cross referencing it with typical anechoic measurements. The JBL Control 1 Pro sample speaker that I measured came with the same measurements performed by Klippel and my data correlates well with theirs. Next step is to get a much better speaker that I can get spin data for.
Those of you who are experts in speaker design and measurement will also need to help get this effort going. Figuring out what useful data is and is not is a major challenge. I don't want to post a ton of graphs just for the sake of it.
Other challenges remain which is to fund purchase of budget speakers and members sending theirs in for review. I have partially solved the shipping cost issue due to kindness of a member allowing me to use his commercial UPS account (for shipping things back). I also have access to some local speaker to test.
Conclusions
"Anything worth doing, is worth doing well." So goes the saying and my position on life. I took a step past the cliff and made a huge investment in this regard to bring order to chaos of speakers. I was encouraged by many member and industry experts that this had to be done. The only way to get it done was me spending the money and worry about getting that back in the future some how. Or not. I don't know if a legacy of testing speaker justifies taking out a chunk out of our retirement budget. You all made me not regret purchasing the Audio Precision analyzer last year. My hope is that you will be there for this investment as well.
So no more complaining about why we are just measuring electronics when that is a "solved problem." We will be measuring speakers and hopefully a lot of it.
Now please excuse me as I go out to buy some warmer cloths so I can work in our cold garage.
Alternatives have been around forever that both reviewers and DIY speaker builders use. These can approximate the speaker performance. Doing so however is quite a lot of work, often requiring measuring speakers outdoors and on tall polls and such. If one were to produce a handful of measurements a year, and could find someone who will do such manual work for little money, it could be done. I am neither. There are thousands of models and brands of speakers. No way can we remotely characterize a useful subset of speakers someone may want to buy using these methods.
Importantly, there has been excellent research correlating a weighted sum of various speaker measurement angles to listener preference. This work has been standardized in the form of CEA 2034 "spinorama." This research though requires access to anechoic chamber measurement and a hemispheric set of measurements all around the speaker. At Harman this is done with a microphone array and rotating the speaker underneath it. Here is a picture I took while I was there:
A turntable under the speaker rotates it and snapshots are taken where the microphones are. Once finished, you can approximate the sound distribution all around the speaker. Then, by researching what percentage of sound arrives at a listener from direct sound (in front of the axis of the speaker) and reflections, you can come up with a graph that can predict how a speaker can sound in a typical domestic room. As implied, direct sound which is usually the only measurement published makes up a portion of what you hear. You hear tons of reflections from the room and the sum total of these determines the tonality of the speaker.
Due to complexity and cost of the above, only a handful of companies have been deploying these measurements and almost none provide that data to public (they are used in design stage of the speaker).
A few years ago the German company Klippel, know for their excellent analysis tools and systems for drivers tackled this problem using a different system (called "Near Field Scanner" or NFS). Instead of moving a speaker and using an array of microphones, it keeps the speaker stationary and uses a "robotic" system to move the microphone around. It uses to very unique and highly useful methods:
1. It makes measurements in near field meaning right next to the speaker. This is sharply increases the signal to noise ratio of the measurements since sound power drops rapidly with distance. Normally doing so would give you an incorrect measurement of what you will hear when sitting far from the speaker. Special signal processing is used to project far-field (where you may be sitting) from near-field measurements.
2. Using advances math, and making dual set of measurements, it can separate the direct sound from reflected one. In doing so, it gives you anechoic measurement in any ordinary space!
Before explaining more, here is a quick video of the system from construction to use:
Since the system is fully automated after some manual setup, you can test a typical 2-way speaker in about 50 minutes. Once there, you get a wealth of information about the speaker. In the standard setup, it defaults to over 500 points which each measurement taking less than 6 seconds. Here is an example run:
And the same in 3-D space:
Here is a shot showing the impact of room reflections on measurements:
The red curve is what you would normally measure. The blue is the corrected one where the reflections (shown in dashed lines) is removed. This is a little monitor speaker though so no response in bass. If there were more, it would likely pollute the measurements more.
The signal processing error increases with frequency (which can be offset partially by increasing the Hankel function order). Fortunately there is no need to measure the full 20 kHz that way. Using the same technique DIY methods use (i.e. gating the measurements so reflections are discarded) gives us the equivalent anechnic data. That is shown with the vertical dashed orange line.
Back to CEA 2034 Spinorama data, Klippel NFS automatically generates them for you:
You want to have a flat on-axis response ideally (black) and smooth off-axis response. If you did that, then your "early reflections DI" would be a smooth line. It is not here:
Note that I have reversed the colors in the bottom chart. You can ignore the sound power directivity as that has no correlation with listening preference (it includes all reflections and direct sound whether you are likely to hear their contributions or not).
We get cool drill down such as these:
So if your room is not a rectangular box, you can decide if you should or should not care about some reflections.
You can also slice up the reflections vertically and horizontally:
Sometimes direct axis will not be good but if you go up or down (your ear relative to speaker tweeter), you may get more flat response.
You also get a bunch of pretty 3-D graphs like these 3-D balloons at any frequency, direction and distance:
Or plot all the frequencies at once:
You can see that as the frequencies increase, the sound becomes more directional in irregular ways as the directivity index showed us earlier.
Compare this to the data manufacturer provides (which is above average):
And yes, classical measurements like impedance versus frequency and phase, distortion, etc. are all available.
Amir Buys the Klippel Near Field Scanner (NFS)
Lots of goodies, right? Well, it was not all good news. The hardware and software with all the options retails for US $90,000. Add a computer, amplifier, miscellaneous items and you are up in $100,000 range. I started a discussion with Klippel and their very helpful distributor in US, Warkwyn a few months ago. After much discussion and great support from the company include founder Wolfgang Klippel, we arrived at a financial arrangement that would not cause someone declaring me totally insane for purchasing this system!
The package which weighed around 300 pounds arrived by air and cleared customs after Christmas. Unlike the video above, I assembled 99% of the unit myself with just an occasional help from my wife for heavy items. Did not want to impose on her any more than I had already done with this expensive purchase! The assembly took about 3 to 4 days, with 12 to 15 hour days. You may have noticed me doing electronic reviews near midnight for this reason.
In order to be able to measure floor standing consumer speakers, I had to purchase the "high Z" version which means the measurement rig can climb to some 12 feet! And needs a large radius around it to boot. This meant I only had one place to put it which was our garage. With winter here and temps 10 to 15 degrees above freezing. and concrete floor, it was not fun. But I managed through it.
Next was getting the software and hardware to work together to create my first measurement. With everyone being on holiday, I had to resolve most of the problems myself although Klippel support in Germany was excellent.
Typical of pro software made for just a few people to run, it doesn't run by just clicking and hitting go. I had to get used to the interface of Klippel software which is ongoing as I type this. I have now run a dozen measurements to a point where I feel kind of comfortable with the system end to end. There are a lot of degrees of freedom from the robotic system setup to initial configuration and way the software runs to get usable measurements.
Road Ahead
No, I am not ready to measure any speakers yet. I have to get more comfortable with the data being generated by cross referencing it with typical anechoic measurements. The JBL Control 1 Pro sample speaker that I measured came with the same measurements performed by Klippel and my data correlates well with theirs. Next step is to get a much better speaker that I can get spin data for.
Those of you who are experts in speaker design and measurement will also need to help get this effort going. Figuring out what useful data is and is not is a major challenge. I don't want to post a ton of graphs just for the sake of it.
Other challenges remain which is to fund purchase of budget speakers and members sending theirs in for review. I have partially solved the shipping cost issue due to kindness of a member allowing me to use his commercial UPS account (for shipping things back). I also have access to some local speaker to test.
Conclusions
"Anything worth doing, is worth doing well." So goes the saying and my position on life. I took a step past the cliff and made a huge investment in this regard to bring order to chaos of speakers. I was encouraged by many member and industry experts that this had to be done. The only way to get it done was me spending the money and worry about getting that back in the future some how. Or not. I don't know if a legacy of testing speaker justifies taking out a chunk out of our retirement budget. You all made me not regret purchasing the Audio Precision analyzer last year. My hope is that you will be there for this investment as well.
So no more complaining about why we are just measuring electronics when that is a "solved problem." We will be measuring speakers and hopefully a lot of it.
Now please excuse me as I go out to buy some warmer cloths so I can work in our cold garage.