• HifiVoice
• Audio
  • Milestones
    • Inspiration Stage
    • Ceramique Stage
    • Wooden Stage
    • Acoustical Stage
    • Measurement Stage
    • Electrical Stage
      • Explanation
      • Wiring
      • Components
      • Experiences
      • Logbook
    • Financial Stage
    • Information Stage
    • Conclusions
    • Acknowledgements
  • Classic 25
  • iPod
  • DIY DAC

My experiences on filter design

Introduction

For an initial filter, there is a good bunch of theory available to create designs without serious flaws (see the information stage pages for some links). But eventually the path to a final filter design is one where theory and personal preferences are interacting intensively. Initial positive impressions of a new filter may turn out to be flaws afterwards when a wide range of music has been tried. Improvements which might seem dull initially, seem to be required for the following step providing true excitement.  The path to a "good" filter, surviving time and a wide range of music, requires time, patience and a good dosis of self critisism.

Lucky you, or a multi-stage approach

In some cases, you plug together some units, spend a few days on a filter, and things seem to work. In most cases, speaker design doesn't work that easily. In my experiences up till now, I've encountered 3 major phases that you need to tackle, in order to get a proper filter design.

Technically correct. With "technically correct", I mean free from big artifacts like mismatched enclosures, resonances, sibilance, phase distortion in crossover regions, no cone breakups in operating region, etc. Errors should be brought back to acceptable proportions, which requires some trade-offs as each system will have some non-ideal properties (e.g. flat amplitude versus aligned phase) Limited technical means and skills make it hard to find artifacts. The low-distortion Thiel units and the solid cabinets hardly mask any errors, so errors become apparent very quickly. Good measurements are required to be able to construct a technical correct design. Despite the many efforts creating such measurements in-house, all attempts failed. For any new speaker design, I would immediately invest in a good anechoic measurement.

Timbre correct. This implies a design with not too much own characteristics or coloration. Unfortunately, this is not just a matter of optimizing the on-axis response only. There are many ways to accomplish a flat on-axis response, and each of them sounds different due to the vertical radiation, and the different off-axis response and hence the energy being generated. Furthermore, there is also the electrical amplifier-speaker interaction, and the effects of the acoustics. As far as I know, there is no software program or method that helps the designer to synthesize a design supporting those ingredients. In practice, this means that optimizing the timbre is an iterative design approach, where most decisions are being taken by listening experiments, and simulations can be used to check or create the next step. In case of the Thiel units, these are not particular "warm" sounding units. When filtered wrongly, the tweeter sounds blunt and separate, the midrange "quacking, "honky", nasal and resonating, and the woofer thin, metallic and/or wooly. The interplay between midrange and tweeter becomes very harsh and oppressive quickly, and a lot of effort is required to get this range under control (too much is oppressive, too little and the treble goes on its own creating sibilance, and not so many filter topologies allow a good integration). Tweaking with less than 0.5dB on a small frequency range is sometimes what is needed to get things straightened. Though simulation make all these "small-scale" effects visible, these are only suitable for analysis afterwards, and hardly for synthesis. Hence, getting the timbre right is starting with a good initial guess, and in the next stage you can only hope that the filter allows for the tweaks that you need. You need a wide range of music, and some reference headphones, because it is easy to fool yourself. This stage costs a lot of time, and gets intermingled with the other stages after a while.

Musically correct. This is a very difficult point to define, and (therefore?) also very difficult to reach. I can only define it as a speaker that is emotionally involving. So that means "some kind of tuning" is required, that is difficult to quantify. It should not only work for intimate jazz combo CDs, but with all the music that I like. Not an introvert speaker that puts a "romantic curtain" over each CD affecting the live-performance, where for instance a close-miked loudly played trumpet should sound "nasty and hard". A close-miked voice should be in-between the speakers, and not before or after the speakers. Too much voicing to obtain such a pleasing effect, reveals artifacts with other CDs. Especially solo piano or guitar is very revealing, some tones become over emphasized, and miss the harmonic structure of the supporting enclosure of the instrument. So, despite some tuning, the speaker should be technically correct, but there is some additional craftmanship required to make it special. This is a road of a lot of experimentation, frustration, learning, unexpected surprises, and after many months it becomes more and more predictable in terms of "do's and dont's". Loudspeaker design is not a cookbook approach, and in case of the Thiel units you're certainly not in the "lucky you" scenario. You need time, money and a lot of patience, plus the will to learn from previous mistakes.

All of this can be summarized by a nice quote of Mahler: "What is best in music is not to be found within the notes"

Some observations

Though most observations are well-known, they are not published, or can only be found as side remarks in text books about filter design. I'll try to keep this list up-to-date with some of my findings.

- Without baffle-step compensated unit curves, a speaker filter should roughly be attenuated by 1dB/oct, or roughly 10dB from 20Hz to 20kHz. This is common practice with headphone design as well. If you start with baffle-step measured unit curves, from 1-2kHz to 20kHz a 3dB attenuation should be taken into account.

- A high-value choke (e.g. 5.6mH) has already a 2 Ohms resistance at 57Hz. As your impedance peak is around these frequencies, a high value choke will decrease the electrical dampening for your woofer, relying more on the mechanical dampening.

- An pi-shape filter will not work correctly for a perfect voltage source. Imagine a CLC filter. The first C will only draw more current from the amplifier, but will not drop the voltage itself, so the capacitor is useless. Use a T-shape filter instead.

next->