'LGT' Construction Diary

[ShinOBIWAN]

Shin:

You take your measurements indoors right? Do you do anything special with the placement of your speaker and microphone before measuring or do you just rely on the gating to clean everything up? About how long is your gate?

Not normally. Instead for the initial driver linearisation I go outside, usually the only time I do in room measurements is to correct for the room. The LGT's are far far too heavy, even when broke up, for one person to safely carry for any distance and especially when negotiating stairs. So for now I'm doing inroom measurements which is full of problems.

Shuffling the speaker iteself well into the room and away from walls, I tend to measure the drivers with the mic on the axis of each and with a distance of 30-40cm to avoid cancellations yet still minimise the room. I try to avoid any gating or minimal gating if possible and this usually is OK down to around 500hz.

Once each driver has been linearised I set the mic back 1.2m on axis with the tweeter. I play a logsweep with all the corrected drivers working and then begin to level match and check overall performance. It doesn't have to be perfect at this point. The above two steps are best done outside if possible or in a decent sized room with the loudspeaker at or close to the center.

From there the mic is moved to listening position and I check the time alignment of the drivers using logsweeps for each driver in turn but with a specific and deliberate delay between each. These sweeps are overlayed and you can see the delays needed by identifying the impulse peaks for each driver and looking at the time differential between each and comparing this to you deliberate delay you engineered into the logsweeps. That's the driver time aligned at the listening position. Then do another single logsweep with all drivers working and check the results. Step response should be markedly improved and the impulse should be cleaner too.

Still with the mic at the listening position you perform another log sweep with the whole speaker working and with the finalised XO. This forms the basis for the room correction.

 

[ShinOBIWAN]

Thought this might interest folks.


Here's the filter transfer functions for the LGT, crossover types are Transient perfect Neville/Thiele 1st order. Also in the overlay is an actual un-smoothed mid driver measurement after correction. The black line shows the perfect summing of the filters.

The design is 3.5way:

W = 8" upto 200hz
M = 5" 200hz - 800hz
T = Ribbon 2500hz+
M = 5" 200hz - 2500hz
W = 8" upto 200hz

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

Plots showing the impressive improvement after driver correction:

Frequency Response
Red = Filter Response
Green = Corrected Driver response(unsmoothed)
Brown = Uncorrected Driver Response(unsmoothed)

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

Step Response

Red = Filter Step Response
Green = Corrected Driver Step Response(unsmoothed)
Brown = Uncorrected Driver Step Response(unsmoothed)

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

Phase

Green = Filter Phase
Blue = Corrected Driver Phase(unsmoothed)
Black = Uncorrected Driver Phase(unsmoothed)

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

 

[ShinOBIWAN]

To provide a bit more background to the why's for the above, here's a post that I made on another forum whilst discussing EQ:

The filter transfer functions are as follows and these, when summed together, create a flat response(the black line). The filters in this example are Linkwitz Riley 4th order:

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

This next one is showing time domain behaviour of the above. You can clearly see the individual filters are ringing, which is typical of any crossover and increasingly severe with steeper filtering, but when summed together the pulse response in black can be seen to show perfect behaviour due to the invidividual filters ringing cancelling each other out when summed as one.

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

Note: These are not driver responses but the visualisations of the mathematical filters and anything can be made to look perfect with maths! So don't pay much heed here. Just know that if the drivers were perfect then so would the speaker. The good news is we know the crossovers themselves are perfect in the time and amplitude domain, at least until a driver is stuck on the end of them and everything is ruined

So what happens when you put your driver in?

Well take a look at this horrible mess showing a typical midrange and its intended passband:

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

There's three elements here:

Red is the filter tranfer function like the ones we demonstrated above, the ideal and what were aiming for in order for the crossover to work accurately and as close to correct as possible.

Brown is the actual driver response with the filter in place and is gain shifted for a better view and not to clash with other the other elements in the graph. Ideally this should perfectly follow the red line but the truth is it couldn't be further from it!

Black(bold) is the minimum phase correction filter needed to bring the brown(driver response) in line with the red(filter transfer function)

Why the need to correct?

Well if you saw that the crossover filters alone are perfect so we must try to make the drivers follow suit and match them to the filter. Its a balancing act here because over-correction ie. correcting every single defect however small or large will lead to a very strange sound that really only works in one very very small sweetspot. An excellent solution is to smooth the correction filter so as to provide a gentler effect that works on the more severe errors but largely ignores the smaller ones that if corrected wholesale could do more harm than good.

Below is a comparison of without correction in green and with correction in blue. Note that the measurements are unsmoothed so what you see is the unblemished truth at the measurement position. You can clearly see the big improvement in overall shape and you can also see that excess correction hasn't taken place because we aren't seeing over correction of the response which brings its own problems. In the end what we have is a driver response that closely follows the filter and should you apply 1/3 octave smoothing to remove the aliasing present in the measurement you'd see that it almost perfectly follows the filter. Do this for every driver, combine the responses and it brings us one step closer to accurate.

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

Further affirmation in the form of a before(red) and after(blue) step response confirming driver time correction was very effective:

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

 

[gbyleveldt]

WOW! Very informative!! Thanks Shin

 

[uli.brueggemann]

originally posted by ShinOBIWAN:
To provide a bit more background to the why's for the above, here's a post that I made on another forum whilst discussing EQ:

Hi ShinOBIWAN,

I'm happy to see you have understood the basic idea and I truly believe you like it

You also should share your experiences at another user group, you certainly know which I mean. Therefore you should also feel free to advertise looking at this completely interesting thread.

 

[schauki]

Hello!

This is a very very impressive speaker. I can only say, respect an superb work.

Looks really fantastic.

Many grats to you!!

cheers

 

[ShinOBIWAN]

Hi ShinOBIWAN,

I'm happy to see you have understood the basic idea and I truly believe you like it

You also should share your experiences at another user group, you certainly know which I mean. Therefore you should also feel free to advertise looking at this completely interesting thread.

Ah the man himself!

Uli your a genius. The superb software you have created is... well... superb! IMO it has furthered the art immeasurably. It's now possible to correct a loudspeaker to the point where it becomes not the weakest link in the audio chain and instead is overtaken by the all important room and to a lesser extent the recording.

Anyone half interested in creating an excellent sounding speaker, and in particular an accurate reference monitor, would do well to seek out Uli's Acourate program.

I am preparing a PDF document that covers the latest advances in PC crossover technologies including sections dedicated to Acourate and Thunau's Frequency Allocator program. Both of these are the leaders in their respective categories IMO. Hopefully it will help other users and people who are intrigued by this approach and would like to try it out. There's a lot of information out there but it seems spread all over the place with a lot of searching required to find the right answers to any question one might have. This is really targeted at those people who tread this path for the first time where failure is highly likely because of the perceived complexity, sadly that also means they miss out on a potential first class audio experience as well.

Once finished I'll put it up on the forums.

 

[ShinOBIWAN]

Many grats to you!!

Thank you Schauki

 

[ShinOBIWAN]

Would anyone local be willing to come and give me a hand getting these sounding right? Now that the initially wow factor has worn off I can hear problems. In terms of getting the sound right, this design is the toughest I've done so far. Initially I thought "yeah, the problems will be worked out with time and experimentation" but its been a good while now and I'm starting to think it might never come together or at least not with my skill set alone. It has crossed my mind that the design itself is compromised rather than pointing the blame at the drivers. So what I really need is another pair of ears and preferably someone with experience who has a good ear. I find myself listening so much that I don't have the greatest perspective any more.

Maybe I should go out and listen to some high end kit so as to get some of that perspective back.

Main problems are a thin upper midrange and a treble that can exacerbate this problem. There are other problems but this is the one that immediately sticks out and is particularly noticeable with female vocals which sound thin and horrible at times.

I can post the full range of measurements if that will help folks but they could be considered very good already so I don't think making the speaker more technically correct is going to help much but please let me know if you think me wrong here, at this point I'm willing to try anything.

Any questions that you have that will lead to results, I will do my best to accomodate.

Thanks

 

[Simon1972]

Ant, i'm just up the road in sheffield and wiling to lend a hand, i do a lot of compartive listening to other peoples rigs and have recently been shopping for a new set of speakers to replace my Kef ref 203's, so have been listening to some 'daddies' in the 3-5k range.

i could even whack the boys in the boot and bring them over to compare, they aren't the most dynamic speaker but they resolve with a beautiful balance and tone and dig out all the detail you could want.

Also i work from home so can bin off a day pretty much whenever i want.

 

[MJL21193]

Hi Ant,
Unfortunately, my Gulfstream G550 is in the shop for a tune-up. Otherwize, I'd be over there to give a hand in the morning.

(yeah, like my ears are good )

 

[ShinOBIWAN]

Ant, i'm just up the road in sheffield and wiling to lend a hand, i do a lot of compartive listening to other peoples rigs and have recently been shopping for a new set of speakers to replace my Kef ref 203's, so have been listening to some 'daddies' in the 3-5k range.

i could even whack the boys in the boot and bring them over to compare, they aren't the most dynamic speaker but they resolve with a beautiful balance and tone and dig out all the detail you could want.

Also i work from home so can bin off a day pretty much whenever i want.

Thanks Simon really appreciate this. With all that listening it sounds like right now your ears are suitably critical too which is just what the speaker needs.

Can you email me z e r o e x _ 1 5 (hat) h o t m a i l . c o m (I tried to make that spam bot proof the spaces need removing.)

We can arrange from there.

 

[Salas]

May I suggest a little experiment if you feel like trying it?
Make your extended mid the upper one, and take as measurement axis the midway point between it and the RAAL at 2m. Pursue a 2dB/dec slant towards HF from 200Hz for total response @ 2m. See if that helps female vocals.

 

[Hezz]

Shin,

Thanks for the fantastic write up on your crossover design approach. Sorry to hear that you are still finding a small portion of the reproduction not quite were you want it. I hate to give any advice since your project pretty much over shadows anything else that I have seen. The female voice covers roughly 200-1000 Hz so if that is where you think something is lacking it could be the just the timbre of the midrange is not quite right. Or maybe it's the upper crossover doing things in the vocal overtones.

Maybe a little more tweaking of the crossover values? Question: Since the PC crossover doesn't have the same interaction problems as an analog passive crossover wouldn't it be good to put that lower crossover right down at the bottom of the male voice fundamentals at 80-100 Hz.

I still think you should experiment with an aperiodic mid range enclosure to approximate infinite baffle for that midrange driver. To see if you like the sound that it makes. Even though the driver may not be specifically made for that type of enclosure it effects the musical timbre to a large degree. Something most engineers never take into consideration.

Anyway, getting a second opinion may be good since after a while you kind of adapt to the sound but in the end it is you that have to be happy with it. However, my thinking is that your speaker may have so many positive attributes that at first a new listener may not notice any weaknesses because they will be so bowled over by the good things which are better than they are used to hearing.

One more thing. I know you experimented with damping material on the rear of the driver. If that midrange driver has one weakness it is that the motor structure is so big that it becomes a reflective surface for the cone to bounce back an out of phase signal into the cone. I know you said that a thick layer of sound deadening material made the sound too over damped. But I wonder if you put a thin layer of felt on all the rear reflective surfaces of the basket and motor structure if that would help at all. Make it sound more refined without be over damped.

All I can say is put on Jane Monheit and if she doesn't sound perfect then something is not quite right. Love that girls voice.

 

[5th element]

W = 8" upto 200hz
M = 5" 200hz - 800hz
T = Ribbon 2500hz+
M = 5" 200hz - 2500hz
W = 8" up to 200hz

Main problems are a thin upper midrange and a treble that can exacerbate this problem. There are other problems but this is the one that immediately sticks out and is particularly noticeable with female vocals which sound thin and horrible at times.

With this setup I find it odd quite frankly that you're experiencing problems with that area.

Obviously the midrange drivers are whats behind this, in one way or another.

For some perspective the lowest note a soprano is normally called on to sing is a C4, also known as middle C and has a frequency of 260hz. For pop and dance this is somewhat 'high' and one could take a G3 as a rough base line. A G3 has a frequency of 195hz.

Top end soprano, highest note, is a C6. Rarely in certain pieces of music, such as Mozart's the magic flute, the soprano is asked to hit notes higher, in this case a F6 corresponding to a frequency of 1396hz. We can assume around an E5 to be roughly the max a pop or dance singer will hit in what is considered to be 'full voice', = 659hz. Occasionally they go higher, but normally these are not sustained notes.

Quite clearly if we ignore harmonics, the fundamental notes are produced by the midrange. The fundamental is, of course, what will supply the weight to the voice, the harmonics changing the flavour of the voice.

All that aside. We could say, is there perhaps not enough reinforcement from the bass drivers? Well not likely, as they will only have any noticeable effect on the lower end of a womans voice. Not to mention, this is three way, and turning the bass up will make it too bass heavy.

Tweeter. By nature this thing is pretty flat, has low distortion and certainly no issues operating at 2500hz. Also has almost zero impact on the fundamental notes. This is not to say it wont excite problems further below mind you, as you have pointed out; the harmonics are still adding to the sound and will put additional stress on a trouble region.

Now to the midrange drivers themselves. Your arrangement of the midrange operating as a 0.5 automatically provides full BSC and could (if both midrange drivers are driven equally) make it sound a little thick, as you'd give it the full 6dB compensation when only 3dB is needed. Quite the opposite of it sounding 'thin', but as this is fully active anythings possible.

I take it you have experimented with increasing the level of the BSC midrange? If anything, this would provide the 'warmth' to help with the thin sound. Thiness in a voice, to me, means it lacks body, which means its harmonic heavy, ie, not enough fundamental in there, of which most will be reinforced with the BSC mid.

Many folk have commented that they do not like a completely flat loudspeaker, that they find it fatiguing, Linkwitz himself makes it no secret that his loudspeakers have a notch dialed in by a few dB around the 1-3khz region to cool them off. Have you tried this? I know completely flat, to me, starts to sound tiring quite quickly, even if it is very engaging in the first few minutes.

As the midrange is quite capable of playing much higher up, you could try it as an experiment. Cross steeply over to the tweeter at say 4-5khz. This will reduce any effect of driver interactions around the important 1-3khz area. If the subjective thiness is still there then we are, safe to say, that its in the midrange and isn't an artifact coming from somewhere else. Of course the only issue with doing this is beaming and the power response. By crossing over higher up you may create a dip, off axis, this would naturally help to combat the 'flat loudspeaker' around this area and may perhaps be desireable if it gives you a pleasing tonal balance.

 

[ShinOBIWAN]

Thanks for the replies. It is late now but I will reply tomorrow.

Before I go I have a set of FR graphs showing the individual filtered driver responses and the almost perfect summed response. Clearly loudspeakers don't come much more accurate in technical terms so I don't think this is the problem.

Please note: The bass driver measurement isn't included as I find it near impossible to measure accurately in-room. Instead the original filter is shown. The others are all measurement with 1/24th octave smoothing applied.

Extreme zoomed in FR showing +0.5/-1dB FR performance.
You can see some very minor cancellation between the two mids.

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

Overall picture:

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

I will try to take distortion measurements tomorrow but such data taken in-room isn't to be trusted wholey.

 

[tinitus]

The woofers can cause this, and now that you say that you find it difficult to measure them, I would say that they may very well be the cause - also, you may have a phase issue with this "modified" 0.5way mid
Besides, I have also experienced MTM to have too much upper mid energy and less lowend

 

[wigginjs]

Shin:

How are you protecting the RAAL from your amplifier?

 

[ShinOBIWAN]

Shin:

How are you protecting the RAAL from your amplifier?

A set of spare ribbons

 

[wigginjs]

HAH!