Articles for Audiophiles by Steve Deckert
AUDIO PAPER #022
HOW TO DESIGN A SPEAKER
by Steve Deckert
There is a good reason why most speakers sound like speakers. It's because most designers follow a specific sequence of mistakes that lead to mediocre sounding results.
Let's face it, if you're going to design a do-it-yourself speaker the first and obvious problem is going to be getting the woofer(s) and the cabinet tuned for flat extended frequency response. Some will choose a cabinet and then model different drivers in it, some will select a woofer and then model different cabinets for it. Both will then design the crossover based on the published specs of the drivers, choosing the best looking spot for the crossover frequencies and then design a network to these predetermined ideals.
They will then order the parts and assemble it. Some will be amazed at the best sounding speaker they've heard, some will be a little unhappy and start tweaking the design. This procedure for speaker design is a recipe for mediocrity. Anyone can do it, almost everyone falls into the trap. To reiterate, you can't design your speaker based on published specs. Some of them are usually never accurate, and others too complex to mean anything in your specific situation.
For example, considering polar dispersion plots of tweeters and midrange drivers as you decide what type of crossover and so on is futile. Drivers all have signatures that defy known specifications, but are in fact the sum of those specifications. You may have all known data from NASA grade test gear on a particular driver, but no accurate way to compile that data in your mind so as to hear what the driver will actually sound like. You actually have to listen to it.
Listening to it is the only accurate way to draw a subjective description of it's sound or signature. And of course the front baffle size, shape, material and angle will also change the signature of drivers, especially midranges and tweeters. So even if you have the drivers in your hand and have listened to several frequency sweeps, the signature you determine the driver to have is false because it's not yet in the cabinet. Once you have the driver in the cabinet, the new signature you determine the driver to have is also false because you haven't installed any crossover components yet. Suppose you installed a cap on the tweeter, and determine the drivers new signature, you still have no idea what the final signature will be because you haven't added the midrange, or woofer. Any drivers and crossover components you add to the circuit will affect every other part of the circuit in the way the current and frequencies divide. So in the end, all the math in the world will not predict accurately how a speaker will sound, and if it will image or not. Although if followed will get you close, but also become a limiting factor of how far you can go.
Added to these variables, as you try to first determine the signatures of each driver, is the room acoustics which will affect the perceived result by at least some 40%. And I've almost forgotten to make it clear that the proper way to design a speaker is by identifying the individual signatures of each component and combining them in a way that is complementary. This can only be done by human ears. A tweeter that has a signature you've determined to be a little sharp on the very top end may be less than ideal when measured against itself, however suppose your midrange drivers is a little bloomed out and overly warm sounding. Depending on the overlap and slope of your crossover, the two can create complementing signatures and actually sound great.
The way to design a mediocre speaker is to use all the math you can and model everything from the crossover to the cabinet to create a perfect result.
If you want a great sounding speaker, you can't design one this way, for several reasons. The first is that published specs on drivers are never accurate. You will have to pick the drivers with somewhat of a casual attitude and then buy them. Once you have them, you'll have to measure them yourself and see exactly what the specs actually are. Be prepared to repeat this process. That means you may be setting some of the drivers you've purchased on the shelf and purchasing new or different ones hoping for better luck.
Some basic steps are as follows:
Once you have the drivers in your hands, and after you've measured the Thiele Small parameters of at least the woofers, you are ready to start designing a crossover. Now that you can plainly see the discrepancies between the published specs and actual measured specs, you can be glad you haven't built the crossovers yet. The first step in designing a crossover is becoming familiar with the characteristics and signatures of each driver. To do this, you simply hold it in your left hand and using a frequency generator, start sweeping the frequencies to extremes of each drivers bandwidth and LISTEN.
You will be listening for smoothness of frequency response. Use your ears, not measurement equipment because the signature of each driver will be superimposed on the frequency response yielding unique sonic results that can't be seen in computer modeled or measured plots.
For example, on the woofer, you will find it's FS just by feeling and looking at the woofer while you sweep it. As you sweep, pay close attention to the sound. Sweep very slowly and hunt for peaks and noises. This process should be documented on a piece of paper by listing the results. For example:
30 Hz ~ 300 Hz - very nice
300 Hz ~ 600 Hz - good
600 Hz ~ 2300 Hz -good
This means that around 300 Hz there was a change, either a peak, or a noise of some kind. Another one at 600 Hz and a noticeable roll off after 2300 Hz. Do the same test on your other drivers (mids and tweeters). Circle the Zone of response that had the sweetest sound. When you find a peak or noise, play with the driver by squeezing it, angling it, shaking it, tapping on it, anything you can do without damaging it. The objective simply to see if anything you do changes the characteristics of the peak or noise. This process usually leads to tweaking the drivers before they're ever installed in the cabinet. For example:
Most stamped frame speaker frames resonate or ring at certain frequencies that depending on size and mass hover between 200 Hz and 1200 Hz. Adding damping to the frame can reduce or eliminate these types of frequency response glitches.
If you're using dome tweeters and or midranges, you will have to install them in the cabinet and sweep them again. Expect large changes in frequency response based on the interactions of the baffle refraction. The surface material, i.e.. wood, felt, can largely affect the signature of the dome drivers as well. Remember that signatures are not found in specs, polar response plots, or transient response tests. There is no way to tell from specs the sonic changes that occur between fabric domes and silk, poly, titanium, or phenolic domes.
From drawing notes on paper you can tell the sensitivity of each driver in relation to each other simply by selecting frequencies that overlap and swapping drivers. For example, the midrange may well go up to 10K and the tweeter may well drop down to 2K, so setting the frequency generator around 5K and A/B the mid and tweeter will tell you which one is louder. The armature trained ear can hear 3dB increments, so if you can tell one is louder, the gain was at least 3dB.
Draw yourself a crude response plot on your paper for each driver and as we said, circle the areas that sounded best. This will help you determine where the best crossover frequencies would ideally be. Once you've come up with a plan based on the real data you just measured with your ears taking into account both efficiency and signature, you can rough in a crossover network. Remember, the midrange and tweeter have a large overlap, and both can play the same frequencies, so in the area of overlap, which one has the most pleasant signature? A question only your ears can answer, and one that must be answered before designing the crossover.
Designing the crossover can be a successful experience provided you keep it simple, at least at first. Using your calculator or computer select the components for either a 6 or 12 dB network that approximately will hit the ideal crossover points you want. If your speaker is for high power applications use the 12dB networks, but if your using lower power amps and higher efficiency speakers you have the option of using either one. When your caps and coils arrive, start with the woofer by installing it in the cabinet and sweeping it several times. Now how smooth has the response become, what changes have taken place. If you like to measure things, you may be amazed at how the impedance curve changes once you have the woofer in the cabinet. Playing with temporarily sealed openings in the cabinet will make radical changes in the the impedance curve, as will the final tuning. Assuming your network is either 6 or 12 dB you will have a coil for the woofer. Once you are satisfied that the woofer is tuned to the cabinet you should sweep it several times with the inductor and without. Notice the signature of the woofer changes completely with an inductor in the circuit. If it makes it sound worse and you have unusually flat response you can and should consider throwing it away.
The following must be done with music, not test tones.
In most systems the mids and tweeters are higher sensitivity than woofers and will need to be padded. This can get complicated with a calculator so you're best bet is to assemble a variety of caps and ceramic resistors, and using alligator clips and test leads, clip together the 6 or 12 dB network that you purchased the parts for and hook up the speaker. Play it a low levels and use this first sound as a starting point. With respect to the polarity of the woofer, the midrange and tweeter (other drivers) can be either in phase or out of phase by 180 degrees. If you have more than one driver such as a midrange and tweeter, you should experiment with changing the polarity of each one at a time and listening to what happens. You will notice large shifts in frequency response and presence. The various combinations possible by doing this can yield either a forward or laid back sound. The reason this happens is timing, or phase angle. The phase angle of a speaker changes with frequency, and in the case of the woofer, affected dramatically by box design and tuning frequency.
Remember, at this point you have what looks like a pile of spaghetti on the floor and it's better if you do not know which way is theoretically correct as far as polarity goes. Just listen. Once you have determined which polarity combination sounds best it is time to start balancing the output of each driver in reference to the woofer. This usually involves padding or shifting of crossover points or both and can be quickly accomplished with a variety of 1 ohm to 10 ohm ceramic resistors. The resistors should be used in series with the drivers. (That's mids and highs only - not the woofer(s) This process should be done while listening to the speaker and simply involves swapping out values until you find something you like. Remember that padding a driver changes it's impedance and that means your pre-calculated crossover points will also change. This could be good or bad, so it's nice to have a variety of capacitor values to swap in and out of the circuit as well.
This process can take anywhere from a couple hours to a couple days and when finished, you'll have a really scary pile of test leads and parts connected together in front of your speaker. It may have been difficult to keep track of the circuit as you swap parts in and out, and that's okay. In fact if the audio gods favor you, you will find that when you try to draw the schematic you'll discover that you hooked things up wrong or at least differently than planned. If this happens and the sound is good, you can smile in the knowledge that you're probably on to something.
This brief paper is just an attempt to deprogram do-it-yourself speaker designers. Remember your ears are more expensive than the best test gear, and work better. The reason we favor this design process is because the actual number of real variables is overwhelming. Things like Signatures and coloration in drivers will change your subjective interpretation of frequency response. A good example is the way a poly dust cap can superimpose a dryness in the midrange that is often misinterpreted as better transient response.
For more in depth information about this technique, you can visit the web page where using this process I design a do-it-yourself speaker that you can build and compare with other speakers to demonstrate the results. In fact, if you truly want to graduate to a level of understanding that yields superior results, try this: Design your crossover before you buy your drivers, like you normally would, based on the published specs and other people's comments. Then put it in a drawer somewhere and proceed with the above technique until your finished. When you build the second speaker, install your pre-designed crossover and compare the results. You should find the experience enlightening and if successful a little embarrassing.