Matching horns to compression drivers, a discussion

Mabat, I don't see what the input power or voltage for the response graph might be. I assume it might be the typical 2.83 volts or 1 watt. I'm not sure what I'm supposed to see in these plots. There is the usual horn cutoff visible in the impedance and response graphs. But, without seeing how this response and distortion changes (or doesn't) as power input and SPL's rise, I don't see how you can say this horn is going to operate well over any given range. Many acoustic devices operate nicely at low levels and they all fall apart at some higher level, just a question of where.

I see this a lot in this forum, where people are showing perfect devices with simulations and measurements only at low power levels. Compare this to how Scott Hinson presents a design like the MEH (or Joseph Crowes IMD measures at 115 db). First, he's going to show how well it handles very high SPL's, then we'll be noticing later that a device that can do high SPL without noticeable distortion, sounds quite nice at low levels. This is a very interesting opportunity I think for audiophiles, which is to piggy back on the great amount of innovation by pro-audio driver manufacturers.

In any case, for whatever theory that horn is trying to demonstrate, I think any real evaluation of it's performance must include distortion measurements at high SPL's. Again, see Joseph Crowe's methods. In order to get IMD at 115 db, he is measuring with a simple first order xover to protect the driver and to avoid issues in the region of no acoustic load to the driver. I don't know how he gets a microphone to have that headroom, I imagine it is with a condensor microphone where input gain can be adjusted.
It seems to me that you're jumping from linear to nonlinear effects, willy-nilly. It just doesn't make a lot of sense then. All the linear effects have a big impact on sound quality even at home listening levels. Not so the nonlinear distortion, so I just don't spend a lot of time with it. YMMV, of course.
I don't understand why an interest in nonlinear effects is "willy-nilly". Nonlinear distortion is the worst, most audible kind of distortion. Linear distortion will remain about the same in relative magnitude regardless of volume and is often benign or inaudible. So, for example, if you have H3 distortion at .01% at 85 db and then you have it at .01% at 105 db, then H3 is very low or inaudible and linear within the range of loudness. It has increased, but the increase is linear with respect to SPL.

But, if H3 distortion goes form .01% at 85db to 1% or 2% at 105 db then it has increased by a nonlinear amount and could be very audible and maybe unpleasant. This is often how we perceive distortion. Often people will think the music is too loud and unpleasant when it is really unpleasant due to distortion. Systems that have very low distortion at loud levels can be a dangerous since listeners do not realize they are listening to high SPLs that could damage hearing.

I don't think this is jumping around. I think this is really the central point or main objective of all our efforts to make design decisions and evaluations. It is to find devices that can reproduce music at listening levels without this distortion. All of the simulations and calculations and discussions are about how to accomplish this goal. But, if you never measure this, you will never know if you have achieved your objective.
I do think Joseph Crowe's measurements are more useful than has been characterized here. You many not prefer his horns over others, but they are certainly well enough designed for loading and directivity to compare drivers
They are useful just as Vance Dickason's Voice Coil measurements are useful. The simple point is that if the driver is not measured on the horn you intend to use it on there could be a surprise. It has nothing to do with my own personal feelings, I haven't heard a Yuichi or a Joseph Crowe horn to comment beyond the measurement data.
Nonlinear distortion is the worst, most audible kind of distortion. Linear distortion will remain about the same in relative magnitude regardless of volume and is often benign or inaudible.
I can see why you find Joseph's review measurements to be useful to you. My own experience leads me in a different direction but it is a shame that there hasn't been much progression in distortion audibility metrics since Earl came out with his. Non coherent distortion seems interesting but also has not gained much support.
  • Like
Reactions: 1 user
Fluid, Vance measure's CD's on the horn provided by the MFG. As I stated above that means that when you compare measurements of two different CD's measured by Vance, you may not be able to make a distinction between the CD's performance versus the horn's performance. My contention is that better conclusions about the CD can be made if it is compared to other CD's on the same horn. I believe that it is these relative differences between drivers that are more illuminating that focusing on a single set of data. Of course, you could also compare Vance's measurements to Joseph Crowe's, then you have that CD compared to several others on the same horn, a JCA horn, and measured on the MFG horn by another investigator.

Some people would argue that the CD should be measured on a PWT. And that would be great, but we don't have that data.

I really don't understand what "leads me in a different direction" means. What measurements of distortion are you using instead of harmonic distortion, IMD and the Gedlee GM metric? Yes, it would be nice if there was more work on the audibility of distortion but we can measure IMD and GM now, and Joseph does. And I only really mention his work as a comparison and as the ONLY example I can find of someone measuring IMD. Vance does not.
I've worn out my copy of Vance's cookbook, but he hasn't innovated his measurements at all.
And since Vance doesn't measure IMD distortion at progressive levels of loudness, there is no failure criteria. All of the drivers Vance measures pass the test.

Joseph has developed a pass/fail criteria based on IMD and he has failed drivers. It's very polite and he clearly states that it is against his criteria for audiophile use, but there is a criteria that is now consistently applied over many drives and now many years.

It's really hard to find any criteria like this since there is little consensus on the subjective perception of distortion. Geddes did the research and the outcome was the GM metric, also now used by Joseph. So far accross several drivers it hasn't really revealed much. But, just like the other measurements, it's the long run comparisons over many drivers to reveal whether the measure can assist decisions.

So, okay, you might not like this approach or whatever, but the question remains for someone trying to make decisions, what is your suggestion for how to make this decision? What data, where does it come from, how is it interpreted, are there video's or instructions?

I'm passionate about this point because I have looked high and low for this kind of information, and I have wasted money based on misinformation and the wisdom of crowds, and it just doesn't really exist in a consumable or actionable form other than what I've suggested. I'd love to hear about another solution using current information.
Vance measure's CD's on the horn provided by the MFG. As I stated above that means that when you compare measurements of two different CD's measured by Vance, you may not be able to make a distinction between the CD's performance versus the horn's performance.
That is precisely my point. Measuring different compression drivers on the same horn is not as universal as measuring a dynamic driver on an infinite baffle. If the measurement of the driver is not on the horn you plan to use or at least very similar it is difficult to know exactly what you will get. Of course that does not mean any other comparison is useless it just needs to be taken in context.
I really don't understand what "leads me in a different direction" means.
A euphemism for, I don't agree with your idea of the audibility of distortion in speakers, but I do not want to discuss it, because it always leads to trouble.
Fluid. Thank you, perfectly reasonable. And, I'll just offer, I know my focus here on IMD distortion as an important criteria is in direct contradiction in a way with Geddes research on audibility in that IMD does not distinguish between H2 and higher order distortion so I don't think I can even say the IMD measures correlate well with the GM metric or the recent research by Geddes on audibility for one and then we have the further issue of attempting to correlate these with audibility. Yes, probably just leads to trouble, there's not enough science or published practice for any consensus.

And, I think that's a great point about the difference of a dynamic driver on baffle versus horn. If we go back to Geddes, he's as much said, don't worry about the driver, it's the horn. In his mind, it's that important.

So, to Pano's original question, I might rephrase your summary by saying the only way to do the matching is by simulation or testing on the prospective horn. That's probably good advice for any diyer. It is a bit frustrating to recognize that the only way to really get a good understanding of horns is to get a lot of experience with horns, but that is also a bit freeing as well. Trying to penetrate the acoustic engineering theory is quite painful.
Quick coffee break philosphy:
I think the issue with experience and what sounds what and why extends all loudspeakers not just horns/waveguides and drivers. Basically, everyone needs more or less to experience stuff on their own as we cannot really share auditory experience. It takes some time and effort to realize various aspect of stereo sound depend on many things, many not just from the transducers, or speakers, or even the room but auditory system, and all these make basically single system we then perceive in the end. I like to think that our systems are as good as our listening skill, because even if the speaker was designed and implemented very well, it's still possible to underperform having it in a wrong context (room, acoustics) and / or positioned subotimally, and so on.

So in that sense, it is good to try various things, like various horns and drivers, measure, listen, play around, try to be thoughtful and reason with the stuff to improve listening skill, and after some years a personal experience on all of it has developed some, which likely yields better sounding system than before because now one likely knows what works in their room and which positioning works and so on ;) have fun!:)
  • Like
Reactions: 1 user
Joined 2004
Paid Member
the only way to do the matching is by simulation or testing on the prospective horn.
Well let's hope not! That's the whole point of this thread "How can we find drivers and horns that match before we buy them?"
Of course testing and simulation both work, but with Marco's spreadsheet we at least have a way to determine if the expansion of a horn and driver are a good match or not.

To take a good example, @kevinkr had chosen high quality horns and drivers, both well liked, well reviewed. The Yuichi horn and Radian drivers. But the results were never quite up to what they should have been, not matter how much EQ and positioning he threw at it. He could never get it to sound right. Whas that the horn or the driver? Enter the TAD driver that has an exit flare that closely matches the initial flare of the Yuichi horn. Voila, problem solved. The TAD sounds and measures much better on the Yuichi horn than the Radian ever could.

There are some major differences between the Radian and the TAD driver, an important one being the exit flare rate. The TAD/Yuichi was a good match, the Radian/Yuichi a poor match because of the Radians rapid flare rate. Is one driver inherently higher quality than the other? Maybe - but the flare rate mismatch doesn't give the Radian a chance (on that horn). Perhaps flare rate mismatch plays a very large part of "this driver sounds good, this one doesn't" comments that we have read over the years.

This thread provides a tool to ensure a good flare rate match between CD and horn, which is a large step in the right direction. Get that right first, then worry about the small stuff.
  • Like
Reactions: 1 user
Pano, there is nothing in the thread to suggest or support empirically that it is the CD exit angle to horn throat angle mismatch that is causing some differences on the Yuichi horn. Arez supplied the observation that the Radian 950 has a known 1 khz resonance and the CSD measurement that supports that subjective understanding.

The conclusions from Geddes research can provide a very easy to implement decision heuristic regarding matching the CSD exit angle and the throat profile of a horn though. Geddes basically says that the local or instantaneous rate of change in the profile of the horn is VERY important with respect to distortion and HOM's. There would appear to be a very obvious rule of thumb to suggest that the exit angle of the CSD should be matched to the horn throat profile at the connection. It's intuitive and supported by peer reviewed research.

The suggestion to do this matching by calculating flare rates, etc. starts to look like a design rule based on the impedance loading interaction of the CD throat with the horn, but without really looking at the models already established for that kind of analysis. Maybe a simple comparison of rates could be used, but how it actually relates to the total system impedance is really just conjecture. And there are certainly horn designs that utilize multiple expansion types and rates within the horn profile, like Don Keele's Exponential Conical horns he designed by EV. So, just to say the expansion rate needs to be the same in all sections of a CD horn assembly is somewhat dubious, unless of course you adopt Geddes conclusions about HOM's, which just brings us full circle back to a much simpler explanation and heuristic.

But, if it's the acoustic impedance and response you care about most, you can certainly simulate the total system impedance of the CD and horn including all the important variables including the electrical and mechanical properties of the driver, the rear chamber the throat and the horn. Except, in this case, since the Yuichi horn doesn't follow any of the mathematically prescribed horn expansions, you need a much more sophisticated simulation engine like BEM to model the horn geometry using surface meshes.

If short section flare analysis can be reduced to a general heuristic, you've certainly taken a very niche application, the Yuichi horn, and then used a lot of conjecture and intuition to come to this conclusion. Why not just accept the Geddes research and it's conclusions, for now, until there's a better researched solution.
diyAudio Moderator
Joined 2008
Paid Member
What if a driver produced a plane wave, but for a unique reason that happens while the driver has an angled exit. The question becomes - in what circumstances does the horn need to start at 0 degrees, and when should it start with matching the rate of opening instead... Also, could either be the case on the same device but at different frequencies?
Geddes just simply says continue the profile with the least rapid change to the angle, or in this parlance, rate. So, ideally the horn throat entrance would have an exact match to exit angle of the CD then make any transformation to an expansion, gradually. I think we could think of this as a short PWT which theoretically doesn't change the wave propagation or impedance over a short distance.

Geddes computations all assume the source is a disc and a plane wave. Whether it is, or isn't doesn't really matter, the math requires a starting assumption and whether you start at a disc just in front of the CD phase plug, or the exit, or thereabouts, the following horn profile or CD exit plus horn profile will cause wave shape changes almost impossible to predict, and Geddes only models OS or conical horns.

Geddes then calculates the rest of the wave propagation explicitly using the physics models for acoustic wave propagation and those calculations and his work suggests that large rate changes in profile angle will disrupt orderly progression of the wave down the length of the horn and HOM's will be created. This really isn't new. There's very old published research showing the wave pressure front propagation in a conical horn is a dog's breakfast of disorderly modes not consistent with orderly progression of a plane, spherical or other fancy shape down the length of the horn.

The most practical solution to your question, if you're concerned about small changes in rate, is to use a CD with a very short throat and control all the expansion in the horn. But, this is the crux of the issue, isn't it? Geddes says essentially, don't worry about the CD. With equalization, etc. they're roughly the same. But, many people with a lot of experience believe CD's make a big difference. For, example, I know you have this experience, and I'd be happy to take your recommendation on a "good" CD and exit angle be damned. Some things only come with experience.

So, what's the rule? If I had to guess, and I'm the least experienced person here, I wouldn't use a CD at 20 deg. on a horn that's about 10 deg. or 0 ish deg. even if I was in love with that 20 deg. driver. Why do it?

Or, phrased another way, how often are you faced with a driver decision where there are two drivers, all other criteria matching, where the throat depth and exit angles are only a few degrees different, or more than a few degrees mismatch with horn? If rarely, why bother.
Joined 2004
Paid Member
Unfortunately I have lost the response measurements on the 2380A horns I used to own, but I do not remember there being a large 1kHz hole in the response with the NEO950PB-16. I believed that behavior is related to a horn/driver interaction, but cannot swear to it. Sadly the data that would support my original (and perhaps erroneous) findings is lost.

This is from the attached article and is consistent with what I remember on the JBL horn - I used a comparatively low Q filter in DSP at 2kHz to flatten out the response.


Horn used - from attached article. (Both 760 and 950 are tested on this horn in this article.)


I based my purchase decision on this article and was very happy with these drivers until I put them on a pair of Yuichi A-290 clones.


  • Test Bench Radian Audio Engineering 950PB and 760NEOPB Compression Drivers audioXpress.pdf
    8.8 MB · Views: 20
  • Like
  • Thank You
Reactions: 2 users
Joined 2004
Paid Member
Pano, there is nothing in the thread to suggest or support empirically that it is the CD exit angle to horn throat angle mismatch that is causing some differences on the Yuichi horn.
I'll disagree with that.
So, ideally the horn throat entrance would have an exact match to exit angle of the CD then make any transformation to an expansion, gradually.
Yes, and that is what is discussed earlier in the thread. Match the expansion rate.
The most practical solution to your question, if you're concerned about small changes in rate, is to use a CD with a very short throat and control all the expansion in the horn.
That has been discussed, but is it practical? If you have a driver with a rapid expansion in its throat, what choice do you have in the horn? Maybe if the CD throat is very short it won't make a big difference, I don't know. Is the Radian throat short or medium - certainly not long? It did not work well on the Yuichi, that's for sure.

I have used, owned and heard a lot of horns and waveguides over the past 40+ years. The rapid flare type are not my cup of tea, I prefer a horn with a slower expansion. No matter if the driver matches the flare rate of the short waveguide, they don't sound as smooth or realistic to me as do the deeper horns. But certainly matching the flare rate somewhere has to help, no matter the type of horn.
That review of the Radian 950 looks weird to me. Given the Radian 950 is a 2" exit horn and 4" dia. rated for response by the mfg down to 500 hz, I don't understand the decision to review the CD on a 800 hz. cutoff horn. That's a small horn for a big driver. Maybe it doesn't matter on a waveguide? I don't know, I'm the least to know here, but none of this makes any sense to me.

Vance said "First, I decided to test the 760NEOPB (see Photo 1). However, Radian does not manufacture horns so I used 18 Sound’s Model XR2064, which is a 60° × 40° aluminum horn with a 2” throat entry," This is my point about these reviews and mixing and matching measurements with about a dozen variables changed, then making a causation conclusion. Vance is doing this all over the place in his reviews. Like he just picks up whatever horn is on his desk, then writes up a review about a CD. And, Fluid thinks this information is more useful than a Joseph Crowe review? Just because they're done on a JCA horn? I just don't get it.

The XR2064 is a 800 hz. cutoff horn and of thin die cast aluminum construction. Vance's measurements show breakup and a big increase in distortion and also shows up in CSD plot around 10 khz, but Vance says breakup is a 20k. Okay, but this looks pretty terrible at 10k. Arez confirms this with a CSD plot that looks terrible at 1k and 10k. I'm guessing the TAD drivers don't look like this on any test. Not sure there's really anything else that needs to be said. All the data just points to the 950 being less than audiophile quality by anyone's standards of evaluation. Why start speculation about exit angles and fantasies about wave front shapes?

I don't see how any generalized conclusions can be made from any of this. It's a roulette wheel of experimental design where key variables are changed randomly then a bet is made on Red #15 being a winner. Unless, we just say, if the CSD looks like garbage in your pass band, stay away!
Joined 2004
Paid Member
Yes that is the very point of this thread. One should match the horn and driver, otherwise it’s not a fair test. In an exaggerated case I actually have a pair of BMS coaxial 2” drivers on Altec 1.4” throat horns. If I made any conclusions about the BMS used like that, I’d be laughed at. (Actually the low end is rather good.)

Now that is an extreme case, but I use it to illustrate that a driver/horn mismatch must be taken into account. Otherwise we don’t know the best of either.
  • Thank You
Reactions: 1 user
Yes, exactly. I'm still really curious about this.

I really like the Don Keele CE (Conical Exponential) composite flare horn design for thinking about this. This horn and the EV Constant Directivity horns were the first constant directivity horns, per Kohlbrek. And, Don also published a paper on the optimal mouth size at this time. These papers are bit dated, but I think his approach is instructive for reduction of the question of compatibility matching into generalizations.

Don's design includes 3 distinct sections of the EV horn.

1. Exponential expansion for the first third of the horn. This section is optimized for efficiency and loading. Don arrived at the optimal exponential section by comparing the permutations of purely exponential, conical exponential, mouth size and length. This is mostly about low frequency response or gain and pass band ripple optimization.
2. Conical section for 2nd third. This was designed to achieve directivity objectives.
3. Mouth flare section for last third. The flare rate was increased to reduce the midrange narrowing and lobing seen in multi-cell and other horns. We might also compare this to other treatments of mouth diffraction and increases in apparent size by Le Cleach profiles.

We see some degree of these features in many horns, although not so obvious as distinct flare rate transitions nor treated so explicitly in the design phase as each section solving distinct problems with horn response and directivity.

So, to be so bold, maybe we can think of CD matching in the same way. The CD needs to be a good match to the horn as a lumped impedance compatibility which we'd expect to see in overall efficiency, LF response and power response. Some people might suggest the fundamental resonance of the driver should be below the cutoff, etc. All primarily acoustic impedance matching to the driver.

For simple heuristics, we might then think about exit angles, entrance angles, flare rate compatibility, etc. as directivity and distortion matching. Geddes solutions would appear to focus on this and I believe he has said the LF is better handled by amplification and direct radiating drivers. So, the first case above is not addressed.

So, if we are for example interested in seeing how a driver performs with respect to high frequency extension and distortion, we might want to recognize that we should expect the horn used for testing to have a good or somewhat standard influence on directivity and also distortion. Something like an OSWG might be good for this as that profile is optimized for directivity and avoidance of distortion.

But, if we are interested in seeing response and distortion near cutoff and in the midrange, we might want to see the CD tested on a horn with a typical or standardized cutoff and LF behaviour. Not sure what this might be, but something like a Hypex expansion of T = .7 comes to mind. This could then give a good idea of behaviour at the lower end of the intended band of the driver.

The Radian 950 exhibited problems at both ends of the range, which I suggest should have been obvious from a good measurement. The HF distortion was certainly visible in all measurements considered. And Arez may have shown us that the LF was also problematic in terms of stored energy. But, as discussed so much, I think that example is problematic for generalization.