LCD3 Measurements

[purrin]

 
 

How many LCD3s have you measured? If we're talking variance shouldn't you need to test a lot of LCD3s to report variances?

 
We are talking about variances with the Audeze supplied graphs.
 
 

I'm really loving the graph on the HD800 there.

Tyll says they use laser beams to design and test their headphones.
 
 

Quote:

I think Audeze made a mistake supplying their graphs.  It's got too much smoothing and doesn't show both channels and there's too much variance in the positioning for them to mean anything (according to yall anyways).  Or there is actual variance in the drivers.  Either way, the only thing they contribute is paranoia.  If I paid 2k and got a graph that had a weird valley I'd rip my own nuts off. 
 
I get why they wanted to, because professional speaker monitors come with them (or at least mine did) so it gives them that nice flava flave but they ought to either be meaningful or axed IMO, 

I have to admit that DK's graph with the wierd narrow dip at 9k looks semi-disturbing. I were going to be providing FR graphs to customers, I would have taken a closer look at that one (re-position the headphone or junk the driver) so they (customers) don't rip their own nuts off or freak out.
 
Then again, my own measurements show a similar dip at 12k on my pair, and it sounds just fine, at least with music, but now you got me thinking about running test tones through them...

 

[DarknightDK]

Quote:

 
I have to admit that DK's graph with the wierd narrow dip at 9k looks semi-disturbing. I were going to be providing FR graphs to customers, I would have taken a closer look at that one (re-position the headphone or junk the driver) so they (customers) don't rip their own nuts off or freak out.
 
Then again, my own measurements show a similar dip at 12k on my pair, and it sounds just fine, at least with music, but now you got me thinking about running test tones through them...

I agree with you. In fact, my unit is faulty. Working with Audeze for a replacement.
 

 

[koonhua90]

 
Quote:

I agree with you. In fact, my unit is faulty. Working with Audeze for a replacement.
 

How do you know it's faulty? Did you email Audez'e your frequency response and then they decided to tell you that it is faulty? 
 

 

[purrin]

Quote:

I agree with you. In fact, my unit is faulty. Working with Audeze for a replacement.

Awesome. I'm glad they are taking care of you. I've met Alex and Sankar - great guys.

 

[DarknightDK]

Quote:

 

How do you know it's faulty? Did you email Audez'e your frequency response and then they decided to tell you that it is faulty? 
 

No, it is not about the chart. Some inconsistencies in sound in one driver. Complete faith that Audeze will take care of a replacement. 

 

[purrin]

FR Graph Smoothing

This is an important concern that arnaud brought up and I thought it would be nice to provide everyone who is not familiar some examples. There is a school of thought out there that people can't hear more than 1/3 octave resolution with most music (not test tones.) Personally for FR, I like to see 1/6 octave smoothing (assuming that I know the drivers I'm using don't ring or have resonances.) For CSDs or waterfall plots, I don't believe smoothing should ever be applied because it defeats one of the major purposes of that type of measurement: to find driver resonances.
 
I've stated that excessive smoothing on FR graphs probably wouldn't matter much (in terms of sugar coating) the LCD3's response since they sound so smooth in the upper-midrange and treble regions. Let's take a look. Note that the bass rolls off on my graphs because the measurement method I use which doesn't seal the cups.
 
LCD3 - No Smoothing

 
LCD3 - 1/6 Octave Smoothing

 
LCD3 - 1/3 Octave Smoothing

Oops. I guess arnaud does have sort of a good point. 1/3 octave smoothing does hide that dip around 12kHz. BTW, I'm not really too concerned about it for a few reasons. More on this later.*
 
 
LCD2r1 - No Smoothing

 
 
LCD2r1 - 1/6 Octave Smoothing

 
 
LCD2r1 - 1/3 Octave Smoothing

Look mama! Magic.
 
 
 
* Now going back to the dip on the LCD3 and why I'm not concerned:
 

• I've seen these types of dips now and then on horn speakers and other headphones. I can't seem to hear anything bad and I've confirmed via test tones (although not yet on the LCD3), that these dips aren't audible and are probably measurement artifacts (the odd cancellation caused by pads/enclosure, neutrinos moving faster than light, etc.) 
• Where these dips are a concern is when they are really are not dips, but actually severe peaks. This can be verified via waterfall plots. There a peak at t = 0ms turns into a nasty ridge by t = 0.5ms, then we've got a problem. Most of the time, it's going to be audible. Looking at the LCD3 waterfalls, the 12kHz dip does not appear to be a driver resonance.
• Even if the dips are real, narrow dips are much more benign than peaks. It's just the way our auditory system works.

 

[purrin]

FR Graph Scaling

Another concern arnaud brought up was the scaling with FR graphs. As readers / consumers, we need to be very careful of the ranges used. We also need to be careful of the horizontal and vertical ratios of FR graphs. These things can be adjusted to make things look better than they really are. Let's take a look at the following graphs comparing the HP1000 and LCD3.
 
Audeze LCD3                                                                      Joe Grado HP1000
 
 
At a glance we note the awesomeness of the LCD3. Indeed it's more linear than even the heralded mysterious Joe Grado HP1000, which is known for its neutrality. However, looks can be deceiving.
 
To be continued...
 

 

[purrin]

FR Graph Scaling (continued...)

 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 
 
So what I did was play a trick on you guys. If you were careful, you would have noticed that the LCD3 graph above used a 100db scale and the HP1000 graph used a 50db scale. Now lets put things into proper perspective, using a 50db scale for both.
 
Joe Grado HP1000                                                                  Audeze LCD3
 
This time around, the HP1000 spanks the LCD3 in terms of linearity. This is not to say the LCD3 is a bad headphone. Indeed it is a very good headphone with very smooth response, but it is specially voiced a certain way according to Audeze's vision (sort of like, but opposite to John Grado's vision.)

 

[koonhua90]

^ Purrin, mind commenting on Audez'e vision versus John Grado's vision?

 

[pcf]

Quote:

FR Graph Smoothing

This is a good concern that arnaud brought up and I thought it would be nice to provide everyone who is not familiar some examples. There is a school of thought out there that people can't hear more than 1/3 octave resolution with most music (not test tones.) Personally for FR, I like to see 1/6 octave smoothing (assuming that I know the drivers I'm using don't ring or have resonances.) For CSDs or waterfall plots, I don't believe smoothing should ever be applied because it defeats one of the major purposes of that type of measurement: to find driver resonances.
 
I've stated that excessive smoothing on FR graphs probably wouldn't matter much (in terms of sugar coating) the LCD3's response since they sound so smooth in the upper-midrange and treble regions. Let's take a look. Note that the bass rolls off on my graphs because the measurement method I use which doesn't seal the cups.
 
LCD3 - No Smoothing

 
LCD3 - 1/6 Octave Smoothing

 
LCD3 - 1/3 Octave Smoothing

Oops. I guess arnaud does have sort of a good point. 1/3 octave smoothing does hide that dip around 12kHz. BTW, I'm not really too concerned about it for a few reasons. More on this later.*
 
 
LCD2r1 - No Smoothing

 
 
LCD2r1 - 1/6 Octave Smoothing

 
 
LCD2r1 - 1/3 Octave Smoothing

Look mama! Magic.
 
 
 
* Now going back to the dip on the LCD3 and why I'm not concerned:
 

• I've seen these types of dips now and then on horn speakers and other headphones. I can't seem to hear anything bad and I've confirmed via test tones (although not yet on the LCD3), that these dips aren't audible and are probably measurement artifacts (the odd cancellation caused by pads/enclosure, neutrinos moving faster than light, etc.) 
• Where these dips are a concern is when they are really are not dips, but actually severe peaks. This can be verified via waterfall plots. There a peak at t = 0ms turns into a nasty ridge by t = 0.5ms, then we've got a problem. Most of the time, it's going to be audible. Looking at the LCD3 waterfalls, the 12kHz dip does not appear to be a driver resonance.
• Even if the dips are real, narrow dips are much more benign than peaks. It's just the way our auditory system works.

Hi Purrin,
 
Most of this stuff is beyond me but since my day job involves reading and manipulating graph charts, I have a question:
How is the smoothing done on these FR graphs? I know in my world most of the automatic smoothing is achieved by discarding data at fixed intervals (which I hate). A change of curve type and tangent can also smooth out the curve. What you show here looks like someone has applied some form of B-spline to it. The amplitude around 12kHZ change quite a bit that makes me wonder why anyone would want to read a smoothed out FR graphs? Surely we can see through the "noise" ( in the curve) and just look at the big picture. By the way, I still don't understand how a big dip like that is not audible if there was not some kind of mistake made on the measurement.
Keep up the good work! I'm learning here. 

 
-Paul
 

 

[purrin]

CSD Graph Presentation

Similarly, the same kind of techniques that I mentioned above could be used to generate two very different looking waterfall plots with the exact same set of data. I'll leave you to figure out the details.
 
 
 
In closing, I want everyone to know that I'm not specifically picking on Audeze. I've met Alex and Sanker. These guys are genuine - they have a passion for this stuff. From what I can tell, they are working their asses off and Sankar is a genius. The reason I've used the Audeze headphones as an example is because for some reason, LCD# threads tend to go crazy and have the most misinformed people making up stuff or arriving at strange conclusions based on whatever "measurements" happen to be handy. Everything seen in the last three posts of mine also apply to other manufacturers. (Yes, that includes Sennheiser, who also optimistically tweaks the HD800 FR graphs that they send to their customers.)
 
 
 

 

[purrin]

 
Quote:

^ Purrin, mind commenting on Audez'e vision versus John Grado's vision?

I can't speak for them, but I can surmise.
 
I've heard John Grado listens to vinyl, so maybe that has a lot to do with it. I think he was smart (back in the day) in knowing that a bright and forward character (bumps at 2k, 5k, and 9k) makes his headphones stand out from the crowd. They are fast, snappy, and very articulate sounding, which was directly opposite to other headphones (I'm talking 90s) such as the Beyers and AKGs (of that era). You could get a cheap headphone for $100 and get something similar to his TOTL RS-1. Now I'm not convinced sticking with this formula will work in the 2010s. There's just too much competition now. Head-Fi seems to be causing quite a lot of consumers to pay another $150 to modify their SR80s rather than spending the extra $150 to get SR225s.
 
I feel the Audeze approach is a reaction to a lot of the latest TOTL headphones (T1, HD800, even RS-1) with excessive treble or nasty treble resonances. A lot of modern music, and especially recently released remasters, tends to have a ton of treble content. The LCD2 was not perfect in eliminating the resonances in the upper-midrange to treble, but at least the FR was shelved down so the resonances wouldn't be so evident. The LCD3 has basically perfected this. There are no treble resonances at all except extremely high up in the "air" region. (Personally I am disappointed that they didn't raise the shelf up a little bit like 2db because of this.)
 
Honestly, I just want a manufacturer to release a dynamic headphone that is neutral sounding with the cutting edge resolution of the latest headphones. HD800 is too bright. LCD3 is too dark. My reference for neutrality (and overall excellence) is the UERM IEM. Ultimate ears was not BS'ing when they said they developed this in collaboration with the Capitol Studios engineering team. My preference is for a neutral or very slightly dark sound.
 
Just my 2 cents.

 

[purrin]

Quote:

Hi Purrin,
 
Most of this stuff is beyond me but since my day job involves reading and manipulating graph charts, I have a question:
How is the smoothing done on these FR graphs? I know in my world most of the automatic smoothing is achieved by discarding data at fixed intervals (which I hate). A change of curve type and tangent can also smooth out the curve. What you show here looks like someone has applied some form of B-spline to it. The amplitude around 12kHZ change quite a bit that makes me wonder why anyone would want to read a smoothed out FR graphs? Surely we can see through the "noise" ( in the curve) and just look at the big picture. By the way, I still don't understand how a big dip like that is not audible if there was not some kind of mistake made on the measurement.
Keep up the good work! I'm learning here. 

 
-Paul
 

I surmise the smoothing is done via some sort of median function (of appropriate width 1/3, 1/6 octave) applied to each interval of the data. I know for sure, discarding data is not used. As for why anyone would want to read smoothed graphs - they look a little bit nicer and are a little bit more digestible to the typical consumer (I remember reading somewhere that 1/3 octave is a good way to go because that's about the limit of human resolution when listening to complex music.) Generally 1/3 octave smoothing works well for well behaved systems without ringing or jaggies (I'm talking about mid or high-end speakers or monitors.) For headphones, given what we've seen, I wouldn't apply more than 1/6 octave smoothing or any at all!
 
As for that 12kHz dip on the LCD3, it could be a mistake or artifact from the measurement process. Weird things happen that I cannot explain. I do need to investigate further by playing individual test tones to confirm, or I may disassemble the driver from the enclosure to take a direct measurement. I know it's hard to fathom, but narrow dips really are very hard or impossible to detect while listening to music, especially if they are in the treble.

 

[slwiser]

Great lesson in reading charts.  
 
Note the last one you gave us has two different time scales 0-3ms and 0-6ms which appears to compress and implying that one has less resonance than the other while both are the same data from the same test. One other observation is that the db scales are different, one going much wider than the other. The wider scaling one revealing more resonance where the one with a narrower scale reads as if there were no resonance in the headphone in the 500 to 1k frequency range.  One can be very creative in how data is presented and/or not revealed.

 

[purrin]

Evil huh?