Modifications to Elac Uni-Fi Reference UBR62 Bookshelf Speakers

[terryforsythe]

I have removed the passive crossovers from my Elac Uni-Fi Reference UBR62 Bookshelf Speakers and gone all active. In doing so, I moved the woofer/midrange crossover frequency from 200 Hz to 450 Hz.

Side Note: Elac specifies the woofer/midrange crossover point as being 260 Hz, but my pair measured 200 Hz, which coincidentally is the specified crossover frequency for the Uni-Fi Reference towers. I don't know whether the published specification for UBR62 is wrong or the factory put the wrong crossovers in my speakers. It is water under the bridge now.

I only have a short time listening to them, and I have not yet tuned with Dirac Live. I'll do that after I get the extra cables I need for the individual woofers, midranges and tweeters ran through the wall. Hopefully this week.

Nonetheless, yesterday I did a quick comparison of one all active speaker to the other speaker in stock form (level matched, but not performed as a scientific experiment - just a quick comparison only to give me a quick sense of whether I was proceeding in the right direction). The improvement in the upper bass/lower midrange dynamics is significant. The changes took the modified speaker to the next level. This morning I modified the other speaker and spent some time listening to them. Wow, what an improvement!

I think Elac crossing the midrange at 200 Hz was pushing it a little hard, which can be seen in the harmonic distortion and linear response data Erin posted in his review of these speakers at Erin's Audio Corner. Also, eliminating the passive filter components probably helped with the dynamics.

I'm sure the question will be asked as to why I modified the Elacs instead of just getting better speakers. Long story short is that after our home renovations the boss (a.k.a. my wife) insisted on NO speakers in the family room. I wanted my stereo system operational, so a compromise was needed. The Elac UBR62 speakers in oak/white with the gray grills were the best fit I could find to match the aesthetics of our family room, and the reviews generally were favorable. I purchased them, having never heard them, and a pair of nice looking stands. I set them up while the boss was away. She came home, saw them, and stated “those look OK, I could live with them.” Mission accomplished, at least so I thought. The sound staging and imaging of the UBR62s was impressive, and initially I was pleased. But after spending some time with them, and turning up the volume a bit when the boss was away, the dynamics of the speakers were underwhelming and left me wanting something better. Now I have it, and the boss still is happy.

 

[komhst]

I think you should change wife and then change the speakers too!

 

[terryforsythe]

I think you should change wife and then change the speakers too!

No need to at this point. The speakers sound really good now, but I am still tweaking the DSP parameters and optimizing subwoofer integration.

When I am done, I'll compare them to my KEF LS60s, which I use in my office. Before the changes the Elacs would not have stood a chance at going toe-to-toe with the LS60s. Now, I suspect they will, perhaps better in some respects and worse in others. We'll see. The Elacs do have the advantage of being paired with an 18" servo controlled subwoofer, lots of DSP tuning capabilities and Dirac Live (miniDSP Flex HTx), but the KEFs have better drivers. It will be interesting to see if all of the tuning will make up for the deficit in driver quality.

 

[Rednaxela]

I think Elac crossing the midrange at 200 Hz was pushing it a little hard, which can be seen in the harmonic distortion and linear response data Erin posted in his review of these speakers at Erin's Audio Corner. Also, eliminating the passive filter components probably helped with the dynamics.

Would be nice if you could validate these hypotheses with measurements.

 

[terryforsythe]

I forgot to note that the speaker modifications also include plugging the ports and applying "No Rez" inside the cabinets, both on the walls and the internal braces.

 

[terryforsythe]

Would be nice if you could validate these hypotheses with measurements.

I will post measurements after I finish tuning them.

Before proceeding with this project I measured the harmonic distortion of the woofer and the midrange independently - I disconnected the jumpers between the pairs of binding posts. I confirmed the high level of harmonic distortion Erin measured between 175 Hz and 450 Hz was coming from the midrange. With the new crossover point that definitely will be improved.

I'll see if I can figure out how to use REW or ARTA to measure the response linearity, which was another weak point of the UBR62s. Just listening to the speakers I can hear a significant improvement in the dynamics. It will be interesting to see how much the response linearity improved.

 

[terryforsythe]

Would be nice if you could validate these hypotheses with measurements.

Here is the link to Erin's review: https://www.erinsaudiocorner.com/loudspeakers/elac_ubr62/

His pair also had a 200 Hz crossover point. Take a look at his harmonic distortion and response linearity plots, and compare them to the frequency response plot for the midrange/tweeter. That is what led me to form the hypothesis. My own harmonic distortion testing on the midrange confirmed at least that portion of the hypothesis.

Anyway, I hope to run the speaker cables this week - I need a second pair of hands to guide it while I pull. The conduit is only 1", and I am trying to run quite a few cables in it. After that I'll do the final tune and measurements.

 

[Penelinfi]

The Elacs do have the advantage of being paired with an 18" servo controlled subwoofer,

Wait, how did that get approved in the family room?

 

[terryforsythe]

Wait, how did that get approved in the family room?

It's in the built-in stereo cabinet under the stairs, behind the dark glass.

I pull the speakers out away from the wall when I'm listening to them. I must have been doing that before I took this picture. That is why there is a long cable hanging off of the right speaker. Normally there is just a short cable that runs down the middle of the stand.

 

[terryforsythe]

Below are some modelling results I generated for this project.

The charts pretty much explain the most significant weak point of the UBR62 drivers. Specifically, the tweeter has a lot of diffraction going on centered at 3 kHz and 8 kHz - the directivity index is pretty bad at those frequencies. Moving the midrange/tweeter crossover frequency up from the stock 1.8 kHz to 2.5 kHz helps a little bit with the 3kHz issue, but not much. It is in the high frequencies where I doubt I can get the UBR62s to compete with my LS60s. Elsewhere, though, they may be able to give them a run for their money. Also, Dirac Live did a pretty good job tuning the system for my room before I started the upgrades, and I expect it will do so again.

These charts are generated using VituixCAD, using frequency response and impedance data I measured using ARTA (without grills - with grills the measurements were a little worse in the high frequencies). The frequency response measurements were performed outside with some ambient noise, so they are not perfect. I tried taking measurements in the house with the speakers at least 8 ft. from any walls, and 6ft off of the floor, using gated impulse responses from which to derive the frequency responses. Using that data, though, the correlation of the VituixCAD models with subsequent measurements was not as good as using the data from the outside measurements. Of course, no models are perfect and final in-room tuning needs to be performed.

In my measurements the speaker ports were plugged since I use the speakers with a subwoofer. Also, the woofer has a 100 Hz high-pass filter applied in the DSP modelling. The first two charts are modeled with 24 dB/octave Linkwitz-Riley filters. The bottom two charts are modeled with 48 dB/octave Linkwitz-Riley filters.

In the following charts:

Gray: On-axis frequency response
Orange: Predicted in-room response
Red (bold): Directivity index

Blue: Woofer response
Green: Midrange response
Red: Tweeter response
Light Green: Listening window

The first two charts are with the DSP optimized for in-room response. The last two charts are on-axis frequency response optimized. The DSP parameters used for the first two charts were my starting point for tuning the system.

 

[terryforsythe]

I played some more with the model in VituixCAD. Moving the midrange/tweeter crossover frequency to 3 kHz, using 48 dB/octave slopes, helped the directivity index a little bit at 2.5 kHz, but slightly worse at 3.2 kHz. It is probably a wash.

I was worried about the midrange starting to beam, but the 48 dB/octave filter knocks it down before that becomes a significant issue, though it did negatively affect 3.2 kHz just a bit.

Another plus is that this would improve the power handling capability of the tweeter.

Here are the in-room and on-axis charts:

EDIT: Updated with less aggressive slope - closer to the Harmon Curve. It still will need a bit more tweaking.

 

[terryforsythe]

Regarding the diffraction issues affecting the tweeter's response, based on quick wavelength calculations and using a tape measure on the tweeter and midrange, the 3 kHz diffraction likely is due to the midrange surround and edge. The 8 kHz diffraction likely is due to the structure surrounding the tweeter at the interface with the midrange.

 

[hex168]

Interesting. Oddly, your 450 Hz crossover point would have been cheaper for the manufacturer to have implemented in the first place. Since they are not tyros, I wonder what their reasoning was?

 

[terryforsythe]

Interesting. Oddly, your 450 Hz crossover point would have been cheaper for the manufacturer to have implemented in the first place. Since they are not tyros, I wonder what their reasoning was?

I have a hypothesis on that. Although it sounds counterintuitive, I speculate that it is due to cost and packaging. Here is my speculation:

1. The low pass filter on the woofer uses three laminated steel core inductors, the cheap kind. Crossing over at 200 Hz, there may be a little impact on dynamics, but it probably would not be noticeable to many people. If they were to crossover at 450 Hz, the negative impact on dynamics would have been more noticeable - inductor core losses increase exponentially both with frequency and magnetizing force.

To use air core inductors on the woofer probably would be cost prohibitive, and fitting such crossovers inside the cabinets would present a significant manufacturing challenge. I designed passive filters for the speakers using all air core inductors and fourth order networks, but getting all of the inductors inside the cabinets, and keeping ones close to one another perpendicular (to minimize inductive coupling), looked like it would be tough. There is a vertical brace, parallel to the baffle, that extends the height of the enclosure, probably about 3" or 4" from the back of the cabinet. It would be in the way of mounting crossovers to the wall.

Also, air core inductors have gone way up in price since the last time I bought them nearly 30 years ago. I actually looked into buying a cheap coil winder and magnet wire and winding my own.

Anyway, looking at the size of the components I would need, I ended up leaning toward leaving the crossovers external. My thought was to put them under the stairway next to the side of the stereo cabinet. That meant running more wires to the speakers. I pondered over that for a while and then came to the conclusion that I might as well go all active - much easier to tune. I'll touch on that a bit more below.

2. The midrange has a first order (6 dB/octave) high pass filter - a single 70 uF metalized film capacitor (probably polyester, a.k.a. "Mylar"). That capacitor value appears to have been selected to have the same cutoff frequency as the midrange's natural low end roll off, which is 12 dB/octave. Thus, with a single capacitor they were able to achieve a third order (18 dB/octave) acoustic high pass response.

With regard to tuning the system, this was my first time using ARTA and VituixCAD. Decades ago I used to use LMS and LEAP. With that combination I could generate a fairly accurate model and it did not take much manual tuning of the actual filters built based upon the model. LMS had its own card that fit into a ancient computer ISA slot. I kept a computer stored in my garage just for that. When I went to boot it up, after not having used it in over 20 years, it was dead. I tried replacing the battery on the motherboard, but still no go. I contacted LinearX, the manufacturer of LMS and LEAP, and asked if I could run the LMS board in an external ISA adapter connected to a modern computer via USB. They did not know, and I would have had to spend something like $800 - $1,000 to upgrade my software just to find out.

I found out about ARTA and VituixCAD in this forum. But, having not ever used them myself, I wasn't sure what the results would be. Indeed, after some frustration with ambient noise (wind, neighbor's pool pump, lawnmowers on the golf course, etc.) when I made my outside frequency response measurements, I did some research and found a way to use ARTA to make an impulse response measurement, and generate a frequency response measurement from that by gating out the room reflections. Initially I did that, but the results generated by my model did not correlate well with actual measurements.

I then used the initial outside frequency response measurements in my model, and the correlation improved significantly. Because I am all active, changing the crossovers is merely changing digital parameters. Had I actually constructed passive crossovers based on my initial model, I would have faced much frustration.

ARTA has a lot of great features, but one thing I wish it had was the ability to perform gated frequency response sweeps like LMS does. With LMS I could take frequency response measurements in my living room and the models always were very accurate. I am new to ARTA. Perhaps there is a way to optimize the gated impulse response technique I tried, I don't know.

 

[hex168]

Good speculation. No way to know, of course.

Does REW do what you are looking for?

 

[terryforsythe]

Good speculation. No way to know, of course.

Does REW do what you are looking for?

Not the gated swept frequency response like LMS, at least not that I am aware of.

But, I probably could have used REW data for my models. I made two sets of outside measurements, one set using ARTA and one set using REW.

I ended up using the ARTA data because, using ARTA with an ARTA box I built, it provides the ability to correct for any deviations in my amplifier's frequency response. For convenience, the amplifier I used for testing is a very small Dayton Audio class D amplifier, which is around 10 years old. I don't know how smooth is its frequency response, but probably not very good. A lot of cheap class D amplifiers, especially older ones, are not very linear at the top end of the audio frequency spectrum, and vary significantly based upon the actual load impedance. I don't know whether that is the case with my tiny amplifier, but I figured it would be a safer bet to use the ARTA data having the amplifier correction.

With the speakers hooked up to the actual amplifiers I am using for them, I now am using REW to take the frequency response measurements for tuning. I did some initial tuning, but the actual speaker cables I end up running through the wall may have slightly different impedances. So, I'm going to wait until I have those ran before I really optimize the DSP for the in-room responses.

 

[terryforsythe]

I have some preliminary data to share.

1. The first chart is the frequency response of the left speaker at my listing position, with the grill on, roughly 10 degrees off-axis. I used 105 Hz, 450 Hz and 2.5 kHz crossover frequencies, and tuned the EQ for a response that follows the Harmon Curve. 1/6 octave smoothing is applied.

2. The second chart is the frequency response and harmonic distortion at 86dB/1m (in-room, NOT anechoic). The frequency response is all over the place because the speaker is tuned for the listening position, thus the EQ settings are doing a lot of room compensation for that position. I used a SPL meter to dial the volume in to 86 dB average.

3. The third chart I borrowed from Erin's Audio Corner, in Erin's review of the UBR62. It shows the harmonic distortion of the stock UBR62 at 86dB/1m (quasi-anechoic using Klippel).

Comparing my measurement to Erin's, changing the woofer/midrange crossover frequency from the stock 200 Hz to 450 Hz, the Harmonic Distortion has improved roughly 7 dB from the peak between 200 Hz and 400 Hz; it dropped from -34 dB to -41 dB. I may try changing the woofer/midrange crossover frequency to 500 Hz to see if that will improve it some more.

The stock crossover has a resonant circuit tuned around 7 kHz to bump the tweeter output, but also reduced the harmonic distortion, at that frequency. But, at my listening position that would result in a peak in the frequency response. So, I will leave that alone.

I will post again when I get closer to having everything finalized.

 

[rirelien]

Thanks for the detailed writeups. Interesting! I have wondered for a few years if it would be worthwhile creating a kind of ‘active backpack’ (dsp crossovers and low cost class D amps) to convert some popular bookshelf speakers to active. Your project is a level up.

 

[terryforsythe]

Thanks for the detailed writeups. Interesting! I have wondered for a few years if it would be worthwhile creating a kind of ‘active backpack’ (dsp crossovers and low cost class D amps) to convert some popular bookshelf speakers to active. Your project is a level up.

I think it might be worthwhile, especially for speakers which suffer from poor crossover design.

Class D amps keep getting better, especially low cost units. You might be able to come up with some kits that compete with GR Research's crossover upgrades on price, but also allow customers to tweak the parametric EQs to adjust the speakers for their room.

 

[terryforsythe]

Would be nice if you could validate these hypotheses with measurements.

I will post measurements after I finish tuning them.

I just came across my original frequency response/distortion measurements made on the raw drivers, in the enclosures, using REW in my initial round of outdoor measurements - I did not use them in the modelling (I used the ARTA measurements instead) and forgot where I had saved them (I wasn't even sure that I had not deleted them - I have made many, many measurements and have been purging data I did not use just to try to keep everything organized).

Anyway, the first graph is for the midrange and the second is for the woofer. Note that the dark brown line at the bottom is indicated in REW as being the noise floor. The light brown line at the top is the frequency response.

Midrange:

Woofer:

One interesting feature is that for the midrange, without the crossover, the harmonic distortion peaks at 150 Hz. With the crossover, as shown in Erin's data it peaks around 220 Hz or so. It is highly unlikely that is due to the woofer, per se, because its harmonic distortion is in the noise floor at those frequencies. So, the passive crossover appears to have had some impact on the speaker's harmonic distortion.

Either the parametric EQs I am applying also have a negative impact on the distortion, though to a lesser degree, or the distortion measurements I made today, and shared in the post above, are off by nature of being made in my family room and the impact of room reflections. If the latter is the case, my distortion actually would be much lower than the measurements I posted above if measured anechoically.