Monday, March 16, 2015

Bespoke Aria Super Tweeter Review

I have a super tweeter in my system that crosses over at roughly 8.7KHz. This super tweeter is made by an England-based company (I believe that this company is now defunct) called Bespoke and the model is the Aria, a leaf tweeter in a round plastic cabinet with switch-selectable first-order network crossover points (12KHz, 16KHz, and 20KHz) and switch-selectable attenuation (0db, -3dB, -6dB). I initially bought them because of their high sensitivity (claimed 98dB/W/m) to match with the rest of my system and I personally love the sound of a good ribbon anyway.

The Bespoke Aria Ribbon Super Tweeter
 When I received them, I tried attaching them directly to my system with mixed results: I could hear the promise that these drivers had but was sadly disappointed by the built-in network. Great idea; poorly implemented; mediocre sound. I suspected that a compromise was made with the type of crossover capacitors used to keep the cost down and I was right. After all, how can any manufacturer sell super tweeters for $300/pair retail including switchable networks and use high quality crossover components? Remember that your system is like the links in a chain where the strength of that chain is determined by its weakest link, here the capacitors used in the super tweeter network.

So it was off to the workbench I trotted with brand new drivers in hand to undo the switching features and connect the driver directly to the terminals. While I was at it, I also replaced the internal wiring with some hand-made star-quad wiring (silver-plated OFHC copper, Teflon insulation) I usually use when doing such work and added a bit more internal sound damping. The next step was to measure the performance of the "native" driver (without the built-in network).

Ribbon or Leaf drivers always require some sort of minimalist crossover network to eliminate LF content which could irreparably damage the internal transformer. And with the addition of a 8.2uF Clarity SA capacitor in the signal path (Fc=2.5KHz) I was ready to make a near-field RTA measurement. Placing the microphone centered and on-axis as close to the driver as I could without it touching it, I made the following pink-noise measurement with the REW v5.1 RTA software and my calibrated microphone.

Bespoke Aria Pink Noise, 8.2uF Capacitor

The peak at about 4KHz initially puzzled me however with some quick calculations it appears to be a major cabinet resonance perhaps combined with a mechanical diaphragm/suspension resonance (1128 feet per second/4000 Hz=0.282 feet=3.3384 inches or 86mm; roughly the internal diameter of this tube). Given this resonance, the usable bandpass should be at least an octave above this 4KHz cabinet resonance point (8KHz). BTW, note that the performance of the microphone used to make this measurement was woefully inadequate since it is well known that this type of driver delivers excellent and very linear response to way over 20KHz (i.e., this test also revealed the microphone's upper-limit).

Ignoring the band above 15KHz (the estimated upper limit of the microphone) I realized that such an unexpected peak would require EITHER a very high crossover point with a normal second-order network or higher to minimize the interaction of this peak with the tweeter's bandpass. I opted for a fourth-order network ( more complex, more intolerant to tolerances, but a 360-degree phase shift).

Now comes the task of deciding what style of network design (Butterworth, Bessel, etc.) to use. Many online tools are available for free use to mathematically calculate the component values for crossover points and I encourage you to use any one of these tools for your own designs. One of my favorites is here. While there are words that describe the effects of each type of network on the resulting amplitude and phase of the sound at or near the crossover point, one must consider how this design impacts "voicing" of your driver (that is, the effects of that type of crossover network design on your driver's physical and electro-mechanical properties). Assuming that any network design will operate satisfactorily is an incorrect assumption (you know what they say about assuming anything, right?).

A brief virtue/compromise description of the four most popular crossover designs are presented below:
  • Bessel - VIRTUE: optimally constant group delay in the bandpass (i.e., flat phase, fast settling time). COMPROMISE: Slower initial rate of attenuation beyond the bandpass
  • Butterworth - VIRTUE: optimally flat frequency response and low ripple in the bandpass (i.e., flat amplitude). COMPROMISE: overshoot and ringing
  • Chebyshev - VIRTUE: steepest roll-off rates. COMPROMISE: adds amplitude peaks/dips and even more ringing than Butterworth
  • Linkwitz-Riley - VIRTUE: Uniform amplitude at crossover point. COMPROMISE: adds group delay in bandpass
As you can see, design selection is not a simple task since each type of design has its own virtues and its own compromises; such is the drawback of using passive crossover networks as opposed to multi-amplifier configurations with active crossover networks. The other issue complicating a network design decision is this: What may on appear paper or simulation to be the best choice design could measure exactly as predicted, but in reality the listening results may be absolutely the opposite. In other words, what is sound in theory may not be so in practice. Models can only predict from known variables. Much like forecasting the weather, network-driver, filter-phase or filter-amplitude shifts, and who knows what else dynamic physical interactions are unaccounted for in these theoretical mathematical models. You just have to use them to start and then tweak from there.

For example, here is the near-field measurement of this same driver using a fourth-order Bessel network with a crossover frequency of 8.7KHz. One would predict that not only would the 4KHz peak be resolved because of the steep filter slopes but also that the driver would be uniform above the crossover frequency. However, here modeling and reality deviate as shown next.

Bespoke Aria Pink Noise, Bessel 4 Network, Fc=8.7KHz
While it is true that the network functioned somewhat as predicted, the 4KHz peak is still prominent and now only -5dB down from the crossover frequency. So now what? Obviously this is not going to sound well so what can be done? The process is called "voicing" and another filter is added to tame these unwanted characteristics. Voicing is added when a Zobel network is added to a dynamic driver to tame its rising HF impedance curve and also applied to LF impedance peaks. These filters  "nominalize" the impedance of the driver for the crossover network thereby permitting it to operate as predicted. These same filters can be applied to tame unwanted SPL peaks.

For example, adding a LF L-C filter to the Bespoke can change its SPL characteristics. The graph below shows the change of adding a 0.12mH inductor in series with a 2.0 ohm resistor and then placed across the terminals of the super tweeter (shunt across the terminals).
Bespoke Aria Pink Noise, Bessel 4 Network, Fc=8.7KHz, 0.12mH+2ohm shunt
Well that certainly changed things. The 4KHz peak is now gone but as they say you never get anything for nothing. The SPL is very uniform with the expected HF rise of this type of driver design. However, the acoustic contribution of the super-tweeter into the tweeter's bandpass is now an issue. So what can be said? Passive networks have issues when used in loudspeaker designs.

The other drawback to using passive crossover networks is the interaction between other drivers and crossover components. Remember that adding another driver not only permits that driver to more optimally handle its bandpass, but also introduces other dynamic variables to the existing drivers and network. So by adding a super tweeter and its crossover network to your system, it also changes the performance of the other drivers. In effect, adding a super tweeter can change the entire sound of your entire speaker (including woofer performance), and not just add an extension to the top octave.

So how do you decide what network design to use knowing that it also impacts the operation of the existing system? You must listen to the results and listen very carefully. This means that there is both an art and a science to creating proper overall system design (read you never get anything for nothing; there is always a compromise).

For example, I first tried a Butterworth design and while it sounded pretty good, the super tweeter sounded quite off with an unacceptable upper-midrange glare. I then tried a Bessel design and had similar disappointments where the super tweeter just did not have that pristine clarity I hoped to achieve with this driver. So I was at a crossroads: which is the best compromise? Too much glare or to soft an operation? I chose to eliminate the glare since I find this more fatiguing to listen to over prolonged periods (I listen to my rig a lot). What did I compromise? Bandstop issues.

Another factor I needed to resolve at this same time was sibilance. At some crossover frequencies, sibilance became unacceptably pronounced and less so at others. And of course there is another compromise at play: standard values of capacitors. If I could find a crossover point where an off-the-shelf value was available, the cost of the network components would be kept under control. Inductors can be easily wound or unwound but capacitors must be purchased lower-than-desired values and parallel shunts added to achieve the desired capacitance (if a non-standard value is required). If I selected a point where two or three capacitance values were needed (paralleled), it would raise the cost of the network but it is the only way to get the desired value (another compromise).

My philosophy around building crossover networks is to use lower-cost components to find the neighborhood of the solution and then spring for the big-boy toys in the final design. Iteration, repetition, measurements, and subjective comments all help when making any tweaks to maintain your sanity. Change one thing at a time to keep you from becoming overwhelmed and record all of your data.

Once all experimentation is complete, you should sit down and look at your comments about the various configurations. Nothing is more valuable than analyzing your impressions and correlating what you hear with that design. You typically glean insights to what the configuration is trying to tell you by this analysis rather than by staring at the graphs and numbers. And by all means, proceed slowly, preferably over several days or even weeks. Allow yourself enough time to listen to a variety of music and signal sources to accumulate a final impression.

What did I glean?I observed the rising SPL and correlated it to the rising impedance (Z) of a dynamic driver (similar curves, different Q). I tamed this super-tweeter SPL peak with the same type of L-R network used to tame the rising Z of a dynamic driver. I accomplished this through experimentation and measurement rather than calculations. Finding the right combination took a while but in the end the results paid off big time.

Once the final design took shape, I then replaced the cheap signal-path capacitors with quality versions and viola! Not only were all of the issues resolved, the quality capacitors (here Teflon) allowed the driver to perform at its best. Transient response was improved and neutrality achieved. In other words, the network became transparent and permitted the driver to perform at its optimal peak rather than colored by passive components.

Is the Bespoke Aria a good driver? Yes and no. It needs a lot of help to get it to perform at its best. However, once you do, it is very smooth and transparent. I suspect that other driver choices may be more to your liking since they require less tweaking (voicing). However, their cost is double or more of this driver and by taking the time to tweak it you can gain valuable knowledge about driver and network behavior while adding a high-quality top octave.

BTW, the drivers used in this super-tweeter appear to be available from Monacor in the UK for about GBP 46 each (including VAT).

See also THIS LINK on simple modifications to this speaker.

Yours for higher fidelity,

Philip Rastocny

I do not use ads in this blog to help support my efforts. If you like what you are reading, please remember to reciprocate by purchasing one of my eBooks or through a PayPal donation, My newest title is called Where, oh Where did the Star of Bethlehem Go? It’s an astronomer’s look at what this celestial object may have been, who the "Wise Men" were, and where they came from. Written in an investigative journalism style (like that of the Discovery Channel), it targets one star that has never been considered before and builds a solid case for its candidacy.

My other titles include:

Copyright © 2015 by Philip Rastocny. All rights reserved.

No comments:

Post a Comment

To comment on this blog, you must first be a member. All comments are moderated.

Note: Only a member of this blog may post a comment.