Part Two: Performance in a Large Room and its Relationship to Standardized Measurement Techniques

A new standard set of anechoic curves, that are specified to be printed in a specification sheet for speakers, can predict the performance of a speaker in a listening room. I demonstrate this with measurements of the Revel M106 made in a large listening room. Large rooms lack room modes that can extend the low frequency limit of a mini-monitor. I look at deployment of a subwoofer with the Revel M106 in the large room and discuss problems with mini-monitors that have 5.25 inch woofers.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - Performance in a Large room


Consumer Electronic Association Standard ANSI/CEA-2034

If you look at all literature for the hundreds of speaker brands on sale, you will see that no two spec sheets look alike and almost none provide any real information. The Consumer Electronics Association (CEA), the organization that produces CES, attempted to bring useful, information data to the specification sheets. Along with ANSI, the American National Standards Institute, a standard was created to indicate what information should be presented as well as techniques for making the measurements. This is Consumer Electronics Association Standard Method of Measurement for In-Home Loudspeakers (ANSI/CEA-2034).

Unfortunately I have not seen a single data sheet for consumer speaker systems with the ANSI/CEA-2034 data.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - ANSI/CEA-2034 curve set provided by Harman

Harman supplied the ANSI/CEA-2034 curve set which is shown above.

CEA adapted the curve set above as a standard because no single frequency response graph can predict how a speaker sounds to a listener.

The 5 graphs are best understood with a textbook written by Dr. Toole:

Floyd Toole, “Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms.” (Focal Press, 2008)

Chapters 17 and 18, of the text, cover the measurements and interpretation. A free annotated power point presentation by Dr. Toole covering some of the material is on the Harman website

ANSI/CEA-2034, in part, used work by Harman’s Allan Devantier:

Devantier Allan, “Characterizing the amplitude response of loudspeaker systems”, presented at the AES 113th convention, October 2002, Convention Paper 5638

The paper, written under the leadership of Dr. Toole, is an expansion of Dr. Toole’s work published in 1986, made possible by the use of modern computer systems to process a huge of amount of data.

The curves are generated from 70 measurements made in an anechoic chamber. That has been an excuse for many not to present the standard but a very new, but expensive, Klippel Near Field Scanner has been developed which can make the tests in a standard room. While I do not expect each speaker manufacturer to purchase a Klippel Near Field Scanner, it will find a place in laboratories that can be contracted to do the test.

It is impossible for me to replicate the complete ANSI/CEA-2034 curve set. I have no access to the test facilities or funds to have the tests performed by an independent laboratory, so I present the data Revel sent to me for the M106. The purpose here is to fill in data I could not create, and correlate the measurements I did create.

The single point, on axis, measurement is in black and the ANSI/CEA-2034 listening window (green) closely matches my curve above to 300Hz. This curve above goes down to 20Hz in the big Harman calibrated anechoic chamber, but my quasi-anechoic curves are limited to 300Hz. Since this is an anechoic measurement, the low end rolls off faster than in a room, where it is supported by room modes, as we will see in the next section.

The blue curve is the speaker’s power response. This is a 70-point measurement all around the speaker with different weighting assigned to each and I have no way of replicating it.

The red curve is the first reflection response. The early reflection curve is an attempt to reduce all the data in the horizontal and vertical radiation pattern graphs into one curve. The importance of this curve will be seen shortly, when I look at my in-room measurements which correlate it.

Allan Devantier examined the early reflections from the walls, floors and ceilings of speakers placed into 15 different rooms of varying square area and aspect ratios. Some speakers were symmetrical on the short or long wall and some offset to one side. One room was L shaped and 6 of the 15 room speaker setups were arrayed around a corner (they call this triangular). For each room, they calculated the reflection for speakers in the left, right and center positions, multiplying the total data by a factor of 3 for left, center and right speakers.

Combining the data from the 15 rooms he found:

  • Floor reflections came at 20º down, 30º down, and 40º down on the vertical axis.
  • Ceiling reflections came at 40º up, 50º up, and 60º up on the vertical axis.
  • Front wall early reflections came from the speakers at 0º,+/-10º,+/-20º, and +/-30º off the horizontal axis
  • Side wall early reflections came at radiation angles of +/-40º,+/-50º,+/-60º,+/-70º, and +/-80º off the horizontal axis

Back wall early reflections came at +/-90º,+/-100º,+/-110º,+/-120º,+/-130,º +/-140º,+/-150º,+/-160º,+/-170º, and 180º off the horizontal axis.

The early reflection curve is an average of all this data. My horizontal radiation plots correspond to the front and sidewall reflections. When this is averaged, the entire horizontal plain is included in 10 degree increments. That the wall reflections are coming at all angles is the result of all the different types of rooms in the study. I show the specific wall reflections for one room later in the review. Ideally the early reflection curve should be monotonically decreasing, as should the power responses.

The M106 early reflection curve trends downward as the woofer becomes more directional and the amplitude of the reflections decrease. The early reflection response pops up at the crossover to the tweeter but the waveguide soon gets the curve moving monotonically downward, confirming my measurements.

The tweeter peak in the 4.5kHz – 6.5kHz range is in the early reflection response, as expected, since this result was present in my horizontal radiation and vertical radiation pattern measurements.

I have no space here to go into detail on the DI (directivity index curves) at the bottom of the graphs. These are not direct measurements but are calculated using the listening window, first reflection response and power response. Ideally the DI curves should be monotonically increasing. A small deviation at the woofer-tweeter crossover can be seen. It would be larger without the waveguide.

This black, on axis response, is the only accurate measurement I have of the low end response of the M106. The only way I could approximate this is to go outside and do a ground field measurement. That requires a lot of space between my house and the neighbor’s house which I do not have. I could have performed near field, quasi-anechoic measurement, of the speaker’s low end but this technique produces an artificial bump in the low end.

From the Harman measurements the low end is flat relative to the average value of the midrange. It is -6dB down at 50Hz. In room measurements in a small room I will show that room modes extended the bottom end by about 15Hz.

Large Room Measurements

The pair of graphs below is my measurement of the quasi-anechoic NRC listening window response (repeated from above) and my measurement of the speaker in my large room. To match the quasi-anechoic response, this in-room graph is limited to 300Hz.

The in-room measurement is a 30pt spatial average. A 15-point measurement was made for the right speaker and then repeated for the left speaker. This is different from the in-room curves in my past reviews. In my earlier review, I presented a measurement for a single speaker only. Averaging both speakers is an approach others have been using and publishing. This change reduces the modal response of the room by averaging the differences in the response of the two speakers from room asymmetries.

It is important to note a good looking in-room curve does not tell you if it is a good sounding speaker since you cannot see the individual components in the ANSI/CEA-2034 curve set (listening window, early reflections and power response) that contributes to the response. Some people make the mistake assuming a good in-room curve is all that is required.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - 1 Meter Quasi Anechoic

My large room is a complex shape involving half height walls connecting to larger spaces and a cathedral ceiling (18ft * 20ft -26ft * 12ft -16ft). This is typical of new construction for a family room.

With respect to speaker placement, I measured before I listened. With excellent low cost or free measurement software and a low cost calibrated USB microphone, you have no excuse not to measure. Some electronics with room correction are sold with the required microphone and PC software in the box so you do not have to spend a dime. Measuring first eliminates the long and in my opinion futile method of trying to find the optimal speaker placement by listening to the speaker, moving it, listening again and continuing the process for as long as the speakers owner can tolerate it.

I made single point measurements and observed the response. I then moved the speaker to see if the response was being improved or degraded. As you would expect some placements improved one thing but made another worse. You have 4 variables to play with (front wall distance, side wall distance, distance to the listener’s seat and toe in) so it takes some time to get close to an optimal result. The chance you will get the speaker close to its optimal position using the subjective listen – move – listen approach, given all the placement variables, is small. I recommend this measurement approach for any speaker if you want to hear it at its best. It is not something special required for the M106.

With a good room EQ in the system, one concentrates on response irregularities above 300Hz. Below 300Hz is dominated by the room not the speaker. The room EQ will compensate for changes below 300Hz that occur as the speaker is moved. One does have to be careful to check that the speaker has not been moved to a position where a sharp, deep, dip in the response below 300Hz has occurred. Dips caused by room modes are not correctible by applying electronic equalization.

Once I had a good looking single point measurement, I moved to a 5 point spatial average and finally to 30 point average. The AcoustiSoft RplusD software, which I used for all measurements in this review, does averaging automatically after an additional measurement is made so getting a 30-point average can be completed in less the 15 minutes. Unfortunately RplusD does not support USB microphones. You do not need to do a 30 point average at home. A 5 point average measurement for the final placement of each speaker should be adequate.

Part of the optimization was the toe in offset from the line the directly aligned from the speaker to microphone (listener). Six degrees appeared to yield the flattest results for the M106. Latter I did a more detailed investigate of the toe in alone with the speaker 9ft back. I did the test in the middle of my large room to reduce wall reflections as much as possible. I changed the speaker toe in offset from direct to the microphone (0 degrees) to 20 degrees. In this test I measured at increments of 2 degrees. Moving to 6 – 8 degrees off the axis reduced the 6kHz peak without much attenuation above that frequency. A 3kHz peak starts to grow beyond 8 degrees. This 3kHz peak is also seen in the quasi-anechoic horizontal radiation pattern presented above. This suggests the 6 – 8 degree offset toe in is a good place to start when placing the speaker.

A laser pointer is handy to monitor the angle by measuring were the light lands on the back wall relative to when the speaker angled direct at the listener. The distance from the speaker to the back wall and a calculator will get you the offset angle.

I have no special treatment in this room at early reflection points except the floor which is not carpeted. I placed 3 inches of acoustic foam at the floor reflection points. This makes the measurements closer to what one would get with thick carpet on the floor. I have no carpet in my house which does wonders for my allergies but renders the room useless for reviews. All my subjective listening tests have the floor bounce attenuators in place.

As shown in the plot set above, the final speaker placement is within +/-2dB of the quasi-anechoic response of the listening window of the M106. We are not expecting the listening window to match exactly the in-room response, as we will see shortly.

I provided the final dimensions of the speaker placement in my room on the graph above. This may help provide a starting point if you have a room this large (small room suggestions appear below) but your final, optimal placement may deviate significantly from my placement. As I said at the beginning of this section, all speakers benefit from placement optimization with measurements. For best results in my room with the M106, the distance from the side walls, front walls and speaker to speaker distance were larger than I have found optimal for some other speakers. I will show below I got similar results in a small room where these large offsets from the walls were not possible.

It is a good idea to make a note of the final placement including toe in. Someone could bump into the speaker and moves it. You may also want to move the speakers out of the way when not listening. Something that is much harder to do with a floor standing speaker.

The subjective results (below) mirrored the measured performance, as expected, making the extra time in the initial setup work worthwhile.

Full range in-room frequency response

Below is the same in-room measurement as I used in the last section but I expanded the scale to 20Hz to 20kHz. The more ragged response below 300Hz is the room modes and the interaction of the speaker with the walls and floors dependent. The frequency response deviations below 300Hz are room related and not reflective of the performance of the M106.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - Full range in-room frequency response

The large room does not have any significant room modes near 50Hz to bring the low end up. The low end is about -6dB down at 48Hz in the large room.

As I pointed out in the last section, I did not attempt to place the speaker to achieve good results below 300Hz. It turns out I had some luck. Almost all the room response is enhancing the low end. This can be reduced by a good room correction, as we shall see shortly. The absence of any significant nulls in the frequency response is the lucky part of this measurement. A Room EQ cannot fix a deep null although some units try, making things worse.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - Harman Anechoic CEA-2034


Dr. Toole reviewed my data and produced the curve set above. He traced the early reflection curve and placed it on top of my in-room data. Note the close correspondence. The most significant difference is more energy in the 1kHz – 2kHz. Also note the loss of energy above 7kHz. These same artifacts show up in a measurement I made with a different microphone and measurement system. Note this is the Harman anechoic measurement being confirmed by my in-room measurement. I am not just publishing a Harman anechoic curve and telling you to trust it.

The match to the early reflection curve below 300Hz is not expected since this is the area the response is dominated by the room. Dr. Toole writes:

“The match to your room measurement at LF is fortuitous”.

You can compare my M106 measurement above to the spatial averaged in-room measurements of the Revel F208 presented in the Secrets review of that speaker. Note the very different Y-axis on these curves. My curve spans 27dB and the one below 100dB.

Revel F208 - On the Bench 100dB


The in-room response of the F208 also closely follows the early reflection of the speaker in the ANSI/CEA-2034 curves for the F208. The ANSI/CEA-2034 curves for the F208 have been made public in a paper Harman presented at an AES convention. Unfortunately, you cannot download it free from the web.

The M106 in-room curve shows an abrupt non-monotonic step 2.5kHz. In the F208 curve the step is earlier at 2.0kHz and is smaller in magnitude. The F208 has a 5.25 inch midrange which has less off axis attenuation, for a given angle, than the 6.5 woofer. The F208 has a low 270Hz crossover between its pair of 8 inch woofers and the midrange. This low crossover frequency, in the room dominated area, is hidden by the room response.

If the crossover was a more typical 500Hz in the speaker dominated area, response deviations would be seen. To achieve the low crossover frequency requires more expensive crossover components and the midrange must be robust enough not to introduce distortion around the lowered crossover.

Another difference between the in-room result for the M106 above and the F208 in-room results shown in the Secrets review is the step at 6.5 KHz, which is not seen in the F208 curve. Since the tweeters are the same it is not clear why this occurs. This could be selected components for the tweeter or crossover components for the F208.

The F208 is 2.5X the price of the M106 so you would expect slightly better looking curves although much of what you are paying for is the extended bass range with the addition of two 8 inch woofers in the much larger box.

A recent Harman Labs AES paper suggest the preferred target curve trajectory above 300Hz, in double blind listening tests, that slowly declines above 300Hz:

Olive, Sean, “Listener Preferences for In-Room Loudspeaker and Headphone Target Responses”, presented at the AES 135th convention, October 2013, Convention Paper 8994

The Harman target curve closely matches the in-room response of the F208 shown in the Secrets review more closely than the M106. The F208 tweeter level control could have been reduced when Secrets curve was measured which, of course, yields a downward sloping response curve.

Returning to M106 in my large room, the figure below shows the improvement of the response below 300Hz when the Anthem ARC room correction was placed in the signal path. The slight bump is a result of the target curve I used. The target was set to apply 1.5dB of room gain. Most listeners prefer the sound when some room gain is retained. The value of room gain is a user preference.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - Response with Anthem ARC Room Correction

Adding a Subwoofer – objective considerations and subjective reactions

I used the Anthem ARC room EQ to add a subwoofer. I do not know of another room EQ outside of Harman’s own room EQ system which does the base management function correctly (4th order Linkwitz–Riley crossovers on the subwoofer AND main channels).

The Harman system is a complex mix of hardware system called “SDEC” and a computer program, run during calibration, called ARCOS system. This is a 5 figure system that is run by a Harman certified calibrator. Unique to this system is the ACS (Auto-curve-sum), which optimizes the transitions from the subwoofers to the main speakers through the crossover region.

ACS is a two pass system. ACS does a second run of each main channel with all the subwoofer active at the same time to match delay and perform the final EQ around the crossover. No other dual pass room EQ exists to my knowledge. The system also optimizes multiple subwoofers using the Sound Field Management system they disclosed in an AES paper. One hopes someday Harman will make all this technology affordable.

JBL Synthesis used to sell an external box (BassQ) just to do the optimal Sound Field Management, which optimizes performance of multiple subwoofers, but the BassQ box has been long discontinued for reasons that I do not understand. No equivalent product exists. The Anthem ARC supports one subwoofer. It is possible to use multiple subwoofers without Sound Field Management but it requires careful placement and adjustment of each subwoofers level.

Note that Anthem ARC does not check subwoofer polarity (almost none of them do). You must check the response at 80Hz (band limited pink noise) with a sound level meter. Flip the polarity switch back and forth for the highest reading.

I deployed a NHT 10b subwoofer. This subwoofer is in size (smaller) and price point (lower) than Revel offerings. At 30 pounds it can be easily moved out of the room like the M106 if other family members want to reclaim the room for some reason or just vacuum without worrying about crashing into the sub. The 10b has recently been discontinued. The 10b had a switch to select between anechoic flat or bumped up for car crashes. That is missing from the new one.

Revel has its own subwoofers. The 60-pound B110 was already tested.

The B110 is about 4 times the NHT’s price. Comparing the measurements of the two reviews shows the Revel B110 bottom end is more extended and the THD significantly lower.

I set the Anthem ARC for a 80Hz crossover and kept the 1.5dB of room gain I used when listening without a subwoofer. The M106 was down -6dB at 48Hz in the large room and the NHT subwoofer was down -6dB at 31Hz. 17Hz extension does not sound like a big deal. It is the distortion characteristics and driver compression that make all the difference and it is a big difference.

High distortion and compression issues of the mini-monitor are major issues once the subwoofer is crossed over to the main speaker. Reliable distortion and compression data at individual frequencies below 300Hz requires an anechoic chamber to be accurate. The only available published data can be found in measurements made by the NRC chamber in Canada.

I compared the distortion and compression data for 25 mini-monitors tested with 5.25 inch and 6.5 inch woofers and this made it clear the smaller woofers introduce significantly more distortion and compression above 80Hz at 90dB SPL and 95dB SPL at 2 meters.

You really have no choice but to move the crossover up past 100Hz, perhaps 120Hz for a mini-monitor with a 5.25 inch woofer to get above the worst of the distortion and compression issues. The higher crossover creates all sorts of problems including driver localization and more interaction between the sat and sub at the crossover. Complex, closely spaced room modes and boundary interactions occur in the 100Hz – 200Hz range. You want the main channel, alone, playing in that range with one room EQ section alone to be doing the correction.

Strangely, direct competition to the M106, designed to produce a good-looking ANSI/CEA-2034 plot sets, is disappearing fast. Almost none of the companies that follow NRC design practices are producing mini-monitors with 6.5 inch woofers. It is not clear why this is happening.

The NHT Xd DSP speaker I own (not to be confused with the small NHT 10b subwoofer discussed above) is an integrated satellite – mono subwoofer system that sold for $6000. The NHT Xd DSP speaker has an 8th order crossover at 110Hz to try to keep the 5.25 inch woofer, in the typically small satellite box for a 5.25 inch woofer system, from introducing significant distortion. Despite the 8th order crossover, which drops like a stone at 110Hz, I hear dynamic range limits in the 100Hz – 200Hz range.

NHT came up with kludge solution. NHT added another sub-woofer ($2000 more) and they raised the crossover frequency to 150Hz. At 150Hz it would be easy to localize a mono sub hence the need for one subwoofer for each channel. The satellite and subwoofer had to be as close together as possible. It was a strange looking thing but now the dynamic range issue in the below 160Hz was reduced.

Revel PERFORMA3 M106 2-Way Bookshelf Monitor Loudspeaker - Woofer Port

The Revel M106 with its 6.5 inch woofer and greater internal box volume was a dramatically better solution than the NHT Xd, when used in conjunction with the Anthem ARC room EQ. This despite the subwoofer crossover slope reduced from 8th order for the NHT (48dB / octave) to 4th order (24dB / octave) applied by Anthem ARC (again bass management with a 2nd order LR highpass is not a solution for high fidelity music reproduction when a subwoofer is deployed). More comparisons of the M106 to the NHT Xd can be found in the subjective evaluation section at the end of the review.

I did feel the cone of the woofer of the M106 with my hand and found it moving around significantly with the 80Hz crossover in place on some CDs. I put the supplied port stuffer in and the problem went away. The low frequency range with the port stuffed was reduced, and I would not recommend its use without a subwoofer. The reduction presents no issue with the 4th order 80Hz crossover.

In addition to striving for freedom of compression, we want the sound to be integrated as an instrument moves from tones produced by the mini-monitor to tones produced by the sub-woofer.


Panufnik Concerto for Strings “Hovhaness Trestures”

You cannot see from the front of this CD that it contains the Panufnik Concerto for Strings, Percussion and Tympani. The dynamics of the percussion are huge perhaps as a result of too close microphone placement. Despite this, it is an excellent work to test subwoofer integration. Try track 5. The Tympani is on the left and the un-tuned drums are on the right. Using the spectrogram analysis tool in the Audacity freeware program, I identified the drum tuning in the 70Hz – 100Hz range. That is exactly what I needed for a subwoofer integration test. Lower tuned percussion would wind up in the sub and tell us nothing about integration.

With an 80Hz LR4 (Linkwitz–Riley 4th order) crossover, the mini-monitors are producing all the sound at 100Hz. Mini monitors with 5.25 inch woofers get unhappy with the dynamics as they get stressed and you can hear that in reduced impact. The Revel M106 with the port stuffed showed no strain at all.

The percussion instruments are deployed around the stage. They should image at the speakers. If you push the crossover past 100Hz you will hear localization to the subwoofer. To understand what is going on with all the percussion, you really need to see a performance of the Panufnik. A video is on the internet with a different set of performers.

The CES favorite to demo low bass and dynamics is the first movement of the Vaclav Nelhybel Trittico with Fennell conducting the Dallas Winds (Reference Recording RR-52) and the M106 / Anthem ARC / sub combo did it well enough for an audiophiles jaw to drop. Using spectrographic analysis, one can see that at the beginning of the 1st movement, strong energy is located at 30Hz from the symphonic bass drum being enthusiastically wacked. That tells you about the performance of the subwoofer.

It is, however, the 2nd movement that is the real test with tympani and tuned drums starting the movement in the 60Hz – 80Hz range. Low bass in the same frequency range becomes dominant at 45sec in. A bass clarinet solo centered at 80Hz starts at 1 minute 10 seconds. The M106 had no problem dealing with integrating all this in the range. With the 80Hz crossover, the subwoofer never localized.

Alan Hovhaness

Alan Hovhaness “Symphony #31 for strings “

The double bass sections of the Alan Hovhaness’s Symphony #31 for strings (Crystal CD811) is one of the most difficult recording to get good integration between a mini-monitor and sub-woofer. In the 7th movement, the double bass section often plays fortissimo in the 60Hz – 110Hz range running through the crossover region between the satellite and subwoofer. With the M106, the results were excellent. The best I have heard in this recording outside headphones.

Bass is only a small part of the musical spectrum.

Returning to the Alan Hovhaness’s Symphony #31, Hovhaness writes multiple parts for each section. Looking at the spectrogram, in some sections of the work, fundamentals are coming from 60Hz at the bottom of the double bass range all the way up to the top of the violin at 4kHz at the same time. It is hard to tell how many different parts are written in the score which place different string sections slightly above or below other sections. It looked like at least 12 parts on the spectrograph. With so much going on at the same time in this music, any deviation from neutrality in the response, either anechoic or in-room, will stick out when comparing the M106 at matched level to another speaker. I heard no significant coloration across the full spectrum up to the violin harmonic reaching 10kHz on the Revel M106. The M106 clearly presented the complex interlocking part written for full string orchestra as clearly as I have heard listening through speakers.

To summarize the M106 was superb in the range that it was active when crossed over at 80Hz with a 4th order Linkwitz–Riley network. Note with LR4 LPF set to 80Hz crossover, the speaker is still seeing energy at 70Hz (0.4 down as an absolute number or in decibels -8dB) or even 60Hz (0.25 or in decibels -12dB). I heard no dynamic compression and the subwoofer did not localize.

Part Three will discuss Optimizing the Placement of a Mini-Monitor in a Small Room, listening tests and Conclusions

Preview: A mini-monitor is an ideal solution for a small room. I look at placement issues in depth. As in the large room excellent performance requires a speaker with optimal ANSI/CEA-2034 curve sets. I show the Revel M106 achieves the same performance, matching in-room response curve of the large room. This is a result of it excellent measured performance in the ANSI/CEA-2034 curve set. Lower frequency room modes extend in-room low end response enough that full frequency performance without a subwoofer is possible. Placement of a speaker in small room is critical to getting excellent performance. Listening tests confirm that the engineering of the M106 has been done in service of the music.

This is a three-part article. Here are links to Part One, Part Two, and Part Three