Let's talk Tweeters!

Erik wrote: Let's focus on learning about choices speaker designers make and what that may mean to the end user.

Duke replies: I can't speak for any other designers, but for me speaker design starts with figuring out what the priorities are, and then hopefully finding a way to accomplish them (or come as close as is practical) within whatever constraints are imposed. So the system concept comes first, and then the design specifics follow. Thus I see tweeters (and every other part of a speaker) as just part of a system, and no more than a means to an end, which in high-end home audio is usually something like “a sufficiently convincing perception of listening to live music”.

Erik: First, for any given type of dispersion, speakers need to roll off more or less evenly. You don’t want to be 15 degrees off axis and only hear the mid-range. Ideally the speaker’s dispersion is even across as much of the response as possible, but usually this can only be done starting in the upper bass.

Duke: Yes! The radiation pattern plays a far greater role in what we perceive than most people appreciate. Most of the sound we hear in most home audio systems starts out being dispersed off-axis.

Erik: Next, the wider the dispersion, the more early reflections you may encounter, which can severely affect the frequency response and imaging.

Duke: Agreed. In general early reflections are detrimental, but later reflections can be quite beneficial.

Erik: The very large diaphragms of ESL speakers (Martin Logan/InnerSound, etc.) have fabulous clarity thanks to this effect. They can sound like you have headphones on even with very little room treatment.

Duke: Dipolar electrostats typically have a narrow radiation pattern that minimizes early reflections, and then have a spectrally-correct backwave that reflects off the wall behind the speakers. If the panels are far enough out into the room, that backwave energy is sufficiently late-arriving to be quite beneficial.

Erik: Drivers with different dispersion patterns _may_ also have different rate of decay. Consider a hybrid ESL + cone woofer. The woofer radiates omni-directionally and the wavefront looses energy the fastest, while the ESL panel is a plane wave, with narrow dispersion and looses energy more slowly.

Duke: This would be an example of an extreme radiation pattern discrepancy (narrow-pattern line-source panel combined with quasi-omni point-source woofer). Unfortunately it is not possible to equalize such a system so that the first-arrival sound and the reverberant sound have the same spectral balance (something that would be psychoacoustically desirable – I can explain why if anyone is interested). The best we can hope for is a juggling of tradeoffs that works at our chosen listening distance. In general, the lower the crossover frequency between woofer and panel, the better, as long as we don't overly compromise something else. 

Erik: Two of the most important measurements for me are Cumulative Spectral Decay and compression. The first measures energy storage, or "blur" that a tweeter adds to the sound because it won’t stop fast enough. The second measures how a tweeter’s response changes at different volumes.

Duke: We want to minimize anything that's a source of audible coloration, and in the tweeter region, what's happening in the time domain matters a great deal.

You mentioned compression – great job of paying attention to something that matters, but doesn't show up on a spec sheet! Thermal compression effects not only suck the dynamic life out of the sound, they are often responsible for a speaker system's tonal balance changing with volume level, which can happen if the various drivers have differing thermal compression characteristics (which they usually do).

Erik: Usually when I hear about issues integrating woofers with very light tweeters it's a frequency response issue, and integration with the room issue.

Duke: Note that the “frequency response” we hear is NOT the one in the published curve – that is only the on-axis response, which may not even be our first-arrival sound if we're sitting off-axis. And then the reverberant energy – remember that's usually most of the sound that reaches our ears - may well (and usually does) have a significantly different spectral balance. At the risk of over-simplifying, the “frequency response” that we perceive is a weighted average of the two. So what comes across as a room integration issue may well have its roots in a radiation pattern discontinuity in the crossover region. Thus the room gets blamed for colorations that can be traced back to problems in the radiation pattern, which in turn are (arguably) loudspeaker design deficiencies.

A few additional thoughts:

Imo one incorrect approach to speaker design would be this: Take “the best woofer”and “the best midrange” and “the best tweeter” and combine them using “the best crossover” and put them into “the best enclosure”. This can easily result in an overly expensive speaker whose different parts do not play well together, resulting in a lack of coherence that becomes distracting over time. This would be an example of putting the design specifics first.

Imo one correct approach to speaker design would be this: Choose all of the components based on how well they will work together with each other, with the room, and even with the amplifier towards achieving an intelligently-selected perceptual goal. This would be an example of putting the system concept first.

So my opinion on tweeters is, they are just one part of a system, and imo it's not cost-effective to trade off good system synergy for the sake of using “the best tweeter”. To give an example, I don't think anyone has accused Andrew Jones of using the best tweeters, but those of us who compete with him must admit that he is a master of system synergy.

Given my particular, perhaps somewhat unorthodox set of priorities, tweeters whose radiation patterns change significantly with frequency are not the ideal tools for what I'm trying to do, though they can be made to work reasonably well with careful juggling of tradeoffs. My preference is generally for tweeters that use a particular type of waveguide or horn: Constant-directivity with low diffraction, with the intention of crossing over where the woofer's radiation pattern approximately matches the horn's. System concept first, design specifics second. 

My understanding is that there are two general areas of thermal compression effects. The one that has been studied and documented the best is what we might call long-term compression, which arises from both voice coil heating and magnet heating, the latter inducing a (usually temporary) loss of magnetic strength. Less well studied is what we might call thermal modulation, which as you describe happens very quickly - quick enough to reduce a sudden peak, thereby reducing the dynamic contrast, which in turn reduces the emotional effect of the music because musicians often use dynamics to convey emotion. I would expect thermal modulation to be primarily a voice coil heating phenomenon (with an accompanying increase in resistance)... when the voice coil of a speaker is hit with a 100 watt peak, it’s like touching it with a 100 watt soldering iron. Apparently JBL has patented a voice coil alloy whose resistance doesn’t change much as it heats up, presumably to combat thermal modulation. Others have worked on this too, but far as I know JBL is the first to include it in commercial products (some of their big high-end studio monitors).

A couple of years ago I had an exchange with Floyd Toole about thermal modulation, and he said they had definitely found it with some of their measurements at Harmon, and that in some cases it was pretty bad. He mentioned testing a 3-way speaker whose midrange driver was effectively compressing on peaks by about 7 dB! He’s the one who told me that thermal modulation is an area that needs to be studied more.

My own approach to thermal compression and thermal modulation ended up being the brute force method - high efficiency drivers with big motors and big voice coils that won’t start going non-linear until they reach much higher SPLs than are typical for home audio. This just happened to be a fortuitous side effect of giving radiation pattern control a high priority.

I’d like to learn more about "static compression", but apparently my Googling skills are weak... I couldn’t find a website for speakermeasurements.com, but apparently it is associated with SoundStage. I’m aware of their "Deviation from Linearity" test, is that what you’re referring to? Seems to me it includes thermal effects as well. Anyway kudos to them for running it - Richard C. Heyser used to do something similar back in the day for Audio magazine.

Duke

Thank you for that information and for digging up those examples, Erik!

Interesting that the less expensive speaker beats the more expensive one not only in deviation from linearity, but also in radiation pattern smoothness - look at the 45-60-75 degree off-axis curves. I tip my virtual hat to Paradigm.

Considering how SoundStage makes their Deviation from Linearity measurement, I think it includes not only mechanical effects but thermal ones also, because at 90 dB the speaker is seeing 10 times as much excursion but 100 times as much wattage as at 70 dB. At any rate, given that peaks 20 dB above the average are quite common in recordings that aren’t overly compressed, the deviation from linearity going from 70 to 90 dB may just be the "tip of the iceberg" for real-world effects, if the speaker is driven to average levels higher than 70 dB/1 meter.

My impression is that excursion-related non-linearities decrease gradually with level until they reach a certain point and then they shoot up rapidly.

The same thing is more or less true for thermal effects: It’s not uncommon for a driver to exhibit less than 1 dB of thermal compression at 10% of its AES rated power, often rising to about 2 dB at 50% of its rated power and then maybe 3.5 dB at 100% of its rated power. In other words, in that last doubling of input power, we only get about half as much increase in SPL as we "should have" (1.5 dB instead of 3 dB), Now these ballpark figures come from eyeballing the spec sheets of those few prosound manufacturers who publish compression specs. These numbers are for long-term thermal compression rather than short-term thermal modulation (which has a very rapid onset and then a slower release, unless another peak comes along before the voice coil has had a chance to cool down). My assumption is that there is a correlation between the short-term thermal modulation behavior and the long-term thermal compression behavior.

(Note that the AES rated power is a fairly conservative yardstick; typically the "music program" power rating is double that, and then the "peak" power rating may be double the music program rating, and we probably don't know which of these the manufacturer is using... and real-world, the excursion-limited power handling may be significantly lower than any of these at low frequencies.)

Duke