DSP crossovers

Damn surely poop. That's one way to decipher DSP. But not ours. My two prior articles Breaking Good and Passive struggles, active solutions set the scene to become foreplay for today's climax: digital signal processing as the key to better speaker design.

Our tour guide is again speaker designer Pat McGinty of Meadowlark. "Beyond performing the band-limiting functions achieved by the types of passive circuits we've been using for the past eighty years, we now begin to tackle other exciting problems which were always beyond reach. For most consumers, the array of issues that come along with passive filters probably rarely if ever crosses their minds. The crossover is hidden and assumed to be a perfunctory and benign but necessary component. Those assumptions are incorrect.

As a just arbitrary visual example for a more complex crossover shown on the Vienna Acoustics website, we note 38 parts and associated solder joints which the amplifier 'sees'. The next image is from the same site.

"First, let me get this out of the way: passive filters suck. I know. I designed with them for decades. They perform their function by resonating. They store energy and release it later. The more complex these circuits get, the more of that ringing goes on and the more the signal distorts and smears backwards in time. And, none of that nasty ringing makes your amp's job any easier.

"In addition and equally nasty, passive filters impose added impedance between source (amp) and load (transducer) to decouple the two. This can only be minimized by selecting parts with low impedance but cannot be eliminated entirely. In most midrange and tweeter circuits where output must be lowered to match that of the woofer, shelving resistors add intentional impedance.

"It's a necessary evil that tosses away the electrical damping you'd otherwise enjoy when a transducer couples tightly to an amplifier. That coupling gives the amp firm control over the position of the voice coil which is essential for faithful signal tracking. To oversimplify a little: tight control means snappy starts, changes and stops. Loose control means that the driver's moving mass begins to dominate. It takes a while to get going, then a while to settle down.

"We call these attributes rise and settle times. They're pretty easy to observe on the bench. To make a mechanical analogy, it's the difference between controlling the diaphragm with a stiff stick or a wet noodle.

"Making matters just slightly worse are the peculiar degradations inherent in each type and brand of cap, coil and resistor. When people talk about which capacitor is best, they're really just talking about which is least bad because it adds less distortion and less coloration. The same goes for coils and resistors. I always found that identifying the happiest least-bad combination of parts took quite a bit of effort and expense. I could add a few more points but I think you get it: passive filters suck on at least four counts. Good riddance!

"If you think about it, when audiophiles carefully match an amp to a speaker, a big part of what they're really doing is finding an amp that can contend with a tricky load and still sound nice. What do you say we eliminate that entire silliness and just choose the best amp for the transducer, then couple both directly with a very short stretch of wire?

"That's exactly what DSP allows us to do. With that out of the way, let's look at some of the nifty things we can do with digital signal processing, starting with a much better way to band limit and integrate our drivers. Back when audio first transitioned from all-mechanical Victrola type systems to the amplified systems necessary to execute the slower speed of the LP, no other choice was available to us for bandwidth limiting our multi-way speakers. We've been living with those miserable LCR circuits for something like 80 years.

"But now powerful and cheap digital signal processors have brought us the power to contend with the filter function by rearranging the data stream before converting it to analog to be amplified and, in the process, completely sidestep all of the troubles forced upon us by the old technology. We're free to use modern processing power to make any changes in the data stream that suits our purposes. So yes, very steep cutoffs like 48dB/octave 8th-order slopes. Woofers usually start to cry at the top of their band so we just lop that right off. Tweeters tend to be sensitive to lower frequency input. Push too hard at bottom of the band and you get a raspy bark so we've always had to be quite gentle with our LCR filters. But now we can dare to extend that band downward, then lop it off hard. The result is a wider range of possibilities for marrying the drivers that, in the right hands, brings superior results.

"Plus, we can execute one of the other duties that were undertaken by the old LCR: driver equalization. Usually drivers have an inherent set of nonlinearities, some of which do require attention. The biggie is the typical rising response of most woofers. So we just dial that right out. Of course that is a bit of an oversimplification. The choices one makes depend upon one's objectives. For instance, the solution to making a speaker play extremely loud is utterly different from that of getting the deepest possible bass etc. Whatever the secondary objectives, to me the primary task has always revolved around the human voice. Get that right, get saxophone and piano right and the rest comes easily along. I'm finding that vocal realism is obviously superior to my earlier passive work.

"Making those judgments still relies on the talents of the designer. For me it's an iterative process of comparing alternatives that goes a bit like an eye exam. 'Is this better or this?' You keep choosing the better alternative until you can't make it any better. Then you're done. Now it all happens on a laptop instead of running back and forth across the room to swap out caps, coils and resistors. So you'd expect the new process to speed things right along. Funnily enough, I'm finding that I spend about the same amount of time because now I can probe deeper into the finer points. I'm making better voicing choices and am much more satisfied that I've done my best work. And suppose I eventually do find a better solution? No problem. I just email the new file to my customers.

"Plus, we can make filters for special purposes like optimized low-level listening. Or high output for when you're throwing a dance party. Or voice-forward for TV. Or 'teenager proof' output limiting. Our processors have four presets so a customer could put all four of those programs on his speakers and toggle between them. We post a series of alternative programs in the cloud for easy access by our customers. Try a new program, like it or don't, you can always revert to your preferred one. Savvy customers can watch our (developing) set of tutorials and quickly learn the basics, especially the ins and outs of the infinitely adjustable PEQ. You can try different PEQ on the four presets and switch between them. Fun stuff for a Saturday afternoon and a million times more powerful than fooling around with cables and interconnects.

"And, happy days, now we can simply enforce a clean impulse response. Time only goes one way so that does involve delaying the earliest bits to match up with the latest pieces of the puzzle. So we delay a tweeter by just the right fraction. As one of the few lonely and oft ridiculed adherents to the idea of time coherence, I personally find it hilarious that the very first thing previously time-incoherent designers do when they get their hands on a processor is dial in coherence. They didn't want anything to do with coherence when it was difficult. They scoffed and denied that it even mattered. Now that it's easy.. .I'm claiming the last laugh.

"Back in the day, we had to either slant our baffles back or use stepped baffles to get an alignment in time. Now we do solve that digitally so cabinet design is much simpler and more economic; and we can make better cosmetics. In production, we used to have to tolerate larger unit-to-unit deviations from the target response because hand trimming a passive filter was far too time consuming. Now we can hit the target precisely with just a few clicks. Basically, digital signal processing takes away a whole box of mundane problems so we can pay much more attention to the next and more interesting set of problems. Like how do we improve dynamics? How do we shrink size? How do we get more accurate bass reproduction? How do we get more bass extension? How do we sharpen the focus?  How do we make this integrate with each customer's unique environment? Old enough to remember the death of disco? I predict that's how quickly conventional audio is about to unravel once the consumer realizes how much more value he gets with the next generation of tech."

"One of the trickiest engineering problems we face in conventional speaker design is bass system alignment. The closed box is simplest but fails to deliver as much bottom end as a vented alignment. So vented systems have tended to dominate. For the designer, the vented box is a compromise that must be very carefully weighed. Go for too much extension and you get a sloppy mess. Stick with a tight-sounding choice and you'll get dinged by the press and legions of spec-comparing customers for insufficiently low F3. The problem is an inherent trade-off between extension and ringing. It's a resonant system after all. We're buying extension with ringing. This increased low bass is actually an inaccuracy, a failure to follow the input signal exactly. It's a lie but one we buy into because it sounds warm and foundational.

"During my time with conventional designs, I put up with the extra effort, time, cost and larger box sizes needed to implement transmission lines because the additional resistance that their air column impresses on the diaphragm allowed for pleasing extension with less ringing than an equivalent vented box. Still, in each system we had to deliberately "split the baby" between extension and 'speed'. That's how the old state of the art worked. That trade-off is now gone. We can very simply and straightforwardly make a closed box that does nothing more than optimize the woofer's rise and settle times as the measure of how well output tracks input. Of course just as with closed boxes of the past, the low bass goes away. But now we can simply dial it back in. Now we get superior bass fidelity and extension in a much smaller box simply by applying power and manipulating the signal. You do need a woofer that can take the punishment—read: a more expensive driver—but since we can cut the box size in half or better, I'll happily call that a wash. Maybe it takes a 400 watter on the woofer but who cares? Audiophiles don't know it yet but first-rate amp power is actually pretty cheap. At Meadowlark we have a saying: "Power is cheap. Use lots!".

"Next is the issue of poor band dynamic linearity. Though not often discussed, it is the single biggest shortcoming which makes audio sound not like reality. Nobody talks about it, there's no spec for it and most designers don't even know how to measure it. I'll try not to get too far into the weeds but DL means that every time you double the input, you get double the output; at every frequency across the band. Sorry but regular dynamic linearity looks nothing like the ideal graph shown next. Typical transfer functions droop downward into an irregular shape that's probably worst in the bass. It's why a hifi sounds harder as we goose it. Tonality is changing on the peaks. Fix this and you'll feel a fresh breeze blow through your listening room.

"At first you'd think: 'Okay, let's just adjust the signal in those sagging regions by applying enough power to flatten the dynamic response'. But when you look into the underlying causes of the nonlinearities, you'll see why this won't work. Happily an elegant solution is quick at hand. There are two problems inherent in the way woofers work that sum in a bad way.

"First: at higher amplitudes, more current flows through the voice coil to heat it up. Because heat is a square function of current and impedance rises linearly with temperature, we get long time-constant thermal compression. On top of that the force generated by the motor is not linear. It decreases rapidly as the coil/diaphragm moves away from its center position.

"Ideally, you'd want to keep temps low and excursion in the sweet spot as much as possible. But two inverse square functions work against us. Extension and amplitude each exact an exponential demand for excursion. Ask for both and you have real trouble. You'll be sending the coil into the diminishing force zone for a greater and greater percentage of the time. Not only will you observe diminished linearity in the bass, you'll see the entire band's linearity dragged down with it.

Simply stated: as you demand low bass, you compromise the entire dynamic performance of the woofer. Adding to the trouble is that your tweeter is probably doing just fine. That's why your rig brightens up as you crank it. But there is an easy solution. Let's take a look at the low-frequency behavior of a typical 7" two-way,

"Note the region below 50Hz where you're not really getting any useful bass. Yet you're sending in gobs of current and demand plenty of excursion. Just stop it. Simply snug a high-pass filter up to the woofer's useful output and watch coil temperature and excursion fall off dramatically. You're still getting the same apparent performance without most of the dynamic compression. But you can't do this passively because the bad effects of the increased impedance in the passive filter will kill you. This particular DSP advantage is amongst the most valuable. to design crossovers in the digital domain. The improvement is not subtle. Sparkling dynamics and a sense of ease and readiness are the first things listeners notice when we present our speakers. The net fillet becomes a gentle high pass followed by a boost at the bottom of the useful band. The result is a tiny speaker that sounds very lively and quick and performs like a much larger one."

And there you have it – Pat's solid reasons why DSP crossovers are the future of better loudspeaker design.