LynnOlson

Don't forget the path-length difference between the speakers in the corners and the center speaker, in addition to the 5 foot bass delay in the corner horns. In theory, the most accurate correction would be a HT controller that provided an adjustable delay for the center channel between 5 and 8 mSec. (You need a tape measure to set the delay correctly. At sea level, 1 millisecond is 13.8 inches long.) The bass would then be path-length-synchronized for all three channels, and everything could be wired in-phase. It's true the center mids would then be out of sync by five feet, but then, that's the case for the horns anyway, so the audibility would probably be minor.

Yes, I will. Things still plenty busy around here - the afternoon/evening session with Gary was the one open spot in the last couple of weeks. The American Standard/Trane 14i heat pump/AC is finally installed, and now I'm fine-tuning the settings on the new thermostat. If it ain't one thing, it's another. I have a renewed respect for what the stock Klipsch does - I too would like the Cornwall/Chorus/Forte, or a similar speaker, to join the Heritage line. They're BIG step up from the Heresy, and have a different presentation than the all-horn Klipsch speakers. They're also a good compromise between the mainstream low-efficiency audiophile approach and Altec/JBL school. Looking at marketing demographics, quite a few DHT and vintage-tube enthusiasts would like a Heritage version of the Cornwall/Klipsch/Forte - there are very few speakers on the market today that have good-quality horns from 700Hz on up, a prosound 15" woofer, and a well-balanced autoformer-attenuated crossover.

The crossover description wasn't the easiest thing to follow, so here's a sketch that Gary made for the Chorus II Mod II crossover. As noted before, this crossover is NOT SUITABLE for a stock Klipsch Chorus II; it only works for a time-aligned version with a Fostex tweeter. During the brief time the Klipsch tweeters were working, though, the time-aligned version looked good, and the same smoothness we saw with the Fostex was also present. Lining up the voice coils definitely simplifies the mid-high crossover design, at the expense of esthetics.

Here's the pic that should have appeared with the first post, showing the impulse response of the Mod II version.

OK, here's how the crossover works. Starting with the bass, there's 3.5mH series inductor (stock part), followed by a parallel element. This parallel element consists of a 20uF cap in series with 3 ohm resistor (not stock). The 3-ohm resistor is a fairly critical value, with 2 ohms being too low and 4 ohms too high. The woofer is connected in-phase, as per the stock crossover. The mid horn crossover is the most complex element, with a Universal 3619 replacing the stock Klipsch autoformer. Starting at the amp side of things, there's a series 6uF cap. (The stock series resistor is deleted.) This is followed by a 10 ohm parallel resistor, which parallels the entire autoformer. The amp-side of the circuit is connected to the topmost tap of the autoformer, Tap 5. The circuit ground is connected to Tap 0. The signal emerges from Tap 1 of the autoformer, and before it gets to the mid-horn there is a parallel 1.75mH inductor (stock part). The mid-horn is connected out-of-phase (this is reversed from the stock crossover). The Fostex FT17H tweeter crossover is fairly close to stock, starting with a series 2uF cap (stock value). This is followed by a 160 microhenry parallel inductor (stock). The signal then goes through a 2uF cap in parallel with a 4.7uF cap, or more simply, a single 6.8uF cap (value is not critical). Before the signal gets to the tweeter there is a 50-ohm parallel resistor across the tweeter terminals. Tweeter phase is reversed, same as the stock crossover. Note this crossover DOES NOT WORK for a stock Klipsch Chorus II, or the Mod I version with the Fostex replacing the Klipsch tweeter. It ONLY works for a Chorus II with a Fostex FT17H mounted almost all the way towards the back of the speaker, about 6 inches off the top surface, with some kind of damping material covering the forward top surface of the speaker. By now you are getting the impression that crossovers are NOT a one-size-fits-all exercise, especially if the drivers are moved from the usual locations. For a stock Klipsch, the stock crossover is close to optimum for a speaker intended to placed against a wall or in a corner. When you pull the speaker out into the room, though, the mid and tweeter (mostly the mid) can get a little too hot, about 2dB or so. This is where autoformer taps come in handy.

Here's a different-looking set of data for the Mod II; it shows a 1/6-octave-smoothed response for a 20mSec window, which counts most of the early reflections as well as the direct sound. The participation of the room is obvious in the bass response, which starts to appear with longer window length. Measuring bass response, although easy to model, is actually very difficult. The near-field technique is usually preferred, since that shows total power output into the room, but of course the room (any room) is then excluded from the measurement. In the old days, speakers were measured outdoors suspended from guy wires, but it's generally considered not accurate to measure the bass response of a speaker in a free-field, since room loading dominates the level of the bass. In other words, a speaker designed to be flat in free air is going to be very bass-heavy in any room, even an auditorium, much less a home.

Here's the frequency response of the Mod II using a FFT with a 6mSec window, ending just before the first room reflections. This shows the first-arrival frequency response, which is an important subjective factor to the sound, but not the only factor. The ear/brain also independently processes the room sound, which is a summation of the frequency response into the room (i.e., the first arrival plus all room reflections). Speaker designers usually consider the first-arrival *and* the total response into the room when designing a speaker system, along with other factors such as directivity vs frequency, IM distortion vs frequency, time response, etc. etc. The weighting of these factors (no speaker at any price can do all of these well) is why speakers sound different; each designer picks their own set of priorities. PWK, for example, was famous for putting low IM distortion (and high efficiency) first on the set of priorities. He was almost alone in the industry, but stuck to his guns, to his credit. That design philosophy is why Klipsch sounds the way it does.

Here's the step-response of the Mod I version, with obviously poorer time response. The two different arrival times of the mid-horn and tweeter are clearly visible about 0.8mSec apart. One nice thing about MLSSA, or Sample Champion, or other MLS programs is that you can store the raw *.TIM files on disk and post-process later. I have files I measured back in 1990 on the Quad ELS, CLS II, and other speakers, although my measurement technique back then wasn't quite as good as it now, with a dedicated low-diffraction mike-stand and the ACO Pacific instrumentation microphone.

Here's the step-response data for the Mod II. As you can see, it looks really different than the impulse data, yet is calculated from the same dataset. What's going on is impulse data is heavily weighted towards the high frequencies, while a step-response (similar to a long-duration square-wave) is weighted towards the bass. A "perfect" step-response would rise (or fall, depending on phase) quickly, then gradually decline back to zero (no speaker can reproduce DC, so a return-to-zero is inevitable). Impulse response data, although the native format of MLS systems, can be difficult to interpret, since it mostly shows details that are associated with HF rolloff and things like that. A step-response, as you can see here, makes it much easier to line up the time response of a speaker, if that's your goal.

Here's the Mod I impulse data. True, the crossover was a bit different, but I can assure you a passive crossover can't easily create a discrete 1mSec delay like you see here. The time misalignment is quite visible compared to the first photo ... which by the way, shows a system that is only approximately time-aligned. If we'd spent more time at it (using the step-response data shown in the next photos), we could have dialled it in more exactly, to a few millimeters. But I think the Mod II is within an inch or so of being correct, judging from the impulse photo.

OK, here are the pics promised earlier. The first one is a MLSSA impulse response, with time-aligned Fostex tweeter. The next post will have the earlier Mk I mod, with a somewhat different crossover and the Fostex mounted in the usual tweeter location. Why Fostex instead of the usual Klipsch tweeter? It's not that the Fostex is startlingly better, just that Gary and I are having trouble with the Klipsch tweeter. The data we posted last time for Chorus II with the Klipsch tweeter was no good; or rather, the tweeter was no good, since the replacement diaphragm didn't fit, and was rubbing against the pole-piece. This ruined the measurements. Well, Gary got another set of replacement diaphragms that fit better, but the first one failed in the first half-second of MLSSA testing yesterday. We made lots of checks to make sure no DC was getting through the crossover, no wiring errors were present, and the power amp was free of ultrasonics. No dice, the tweeter must of just been getting ready to fail. Anyway, to make a long story short, we didn't have any good Klipsch tweeters to measure, although for a brief interval the MLSSA data looked good, with nice extension out to 20kHz. So the Fostex tweeters from the last go-round came out of storage and we adjusted the crossover and looked at how it did in a time-aligned setup. So here's the first graph, showing impulse response. The next post will show the conventionally-mounted Fostex tweeter.

Hi all, sorry for the late posting, but things have been busy around here, with a new HVAC system installed today. The twenty-year-old Carrier heat pump that came with the house quit (gee, these things don't last forever?) and finally got the new one installed and running today. The weather here in the Puget Sound region is usually mild, but it has a way of throwing curves at us every now and then. Had another measuring and modifyin session yesterday (I'd like to name this thread Chorus II Part Deux) and liked what we've done so far. Time-aligning the tweeter really chases out those annoying hill-and-dale ripples that seem to plague horn designs. Usually, you twiddle with the phase response (altering crossover slopes, reversing driver phase, etc.) but that seems to just chase the nulls from one part of the spectrum to another. When a design is not responding to the usual crossover tweaks, it's trying to tell you something ... to stop messing with the crossover, and look a little deeper at what's going on. In this case, the mid and tweeter are misaligned by about 1mSec. This is obvious in the time response of the earlier measurements, where you see two peaks, one from the mid, and the one from the tweeter. A millisecond doesn't sound like much, but it's five wavelengths at 5kHz. Looking at it that way, it's a lot. It's a truism in speaker design you can't correct a time error with phase adjustments. Phase is relative to one frequency only, usually the nominal crossover frequency. This implies a phase adjustment that's right for the crossover will be way out an octave above and below. And that's just what we saw on MLSSA - nulls and peaks that moved around, but never going away. Since I'm used to working with direct-radiators, this was a new one for me, ripples that just refused to go away. But I thought a bit about the severe, many-wavelength misalignment of the mid and HF driver. It's not pretty in the time domain - two spikes for the price of one - and has to be having bad effects on the FFT, which is really allergic to reflections of any kind. Even a heavily damped floor reflection at 3mSec (with 20dB of attenuation from many pillows) still creates small ripples in the FFT, so what can two big peaks 1mSec apart be doing? A lot, as it turns out. Once Gary and I moved the tweeter towards the rear of the enclosure (centerline of the tweeter about 6 inches above the cabinet top with several layers of towels on the cabinet top in front of the tweeter), the crossover pretty much dropped into place. The two spikes in the time display merged into one, and the crossover started to behave normally, with the crossover region nice and smooth for a change. In fact, the entire response of the Chorus/Fostex fits into a 5dB window - that's right, plus or minus 2.5dB, with no smoothing at all. MLSSA data to follow, stay tuned.

Might be a while 'till we get to the Khorns, since I don't think Gary or I are planning to buy a new house any time soon. Belle Klipsch or La Scala, yeah, a possibility there. One thing I have to mention is that MLSSA shows all horn systems in an unflattering light. What horns do best is very low IM distortion, and this is difficult to measure with MLSSA, or any system, without an anechoic chamber (big bucks there). MLSSA favors speakers with good time domain performance, and the champs here are electrostats and a few direct-radiators. By comparison, horn speakers don't come off well when you look at MLSSA data - this measurement bias is reflected in the high-end biz in general. So why I am messing around with MLSSA and Chorus at all? Well, I've owned my own MLSSA system since 1991, so I know how to drive the thing (working at Tektronix Spectrum Analyzers and before that, Audionics, helped in this respect). Even though it doesn't conveniently measure IM distortion, I can use it to optimize the time and frequency performance of a horn speaker, keeping in mind that decreasing crossover slopes exacts a price in higher distortion (true with any drivers, not just horns). It's also a little zany and offbeat to apply BBC/KEF/Quad philosophy to a horn speaker, which is usually considered the polar opposite in the speaker-design community. But why not? No reason to keep the sonic balance of the mid-Forties Altec Duplex around forever; it was state of the art then, but we can measure things better today, and can do better with modern crossovers. I have to admit an all-horn system is a lot bigger challenge than the Cornwall/Chorus/Forte series. Horns don't like being "stretched" out of their natural bandwidths (lots of lumps, lots of distortion), and the region where the bass horn quits (300Hz or so) and where the mid horn comes looks tough. Neither horn is really comfortable around 400 to 500 Hz, and that's a critical region of the spectrum. It's the center of the musical power band (most energy), the region where voices come in (bottom of telephone bandwidth), and the few studies that indicate that phase is audible indicate that 500Hz is where it is most audible. The room is starting to get goofy at 300Hz and below, which doesn't help. In fact, unless you have a *really big* anechoic chamber, most measurement techniques require you to switch to nearfield below 300Hz since room effects start to dominate the measurements. If you're getting the idea that 200 to 500 Hz is a rough region for a crossover - much less horn crossovers, which don't "stretch" in any meaningful way - you're right. The Cornwall/Chorus/Forte make all of this a lot easier since getting a 15" or 12" woofer up to 1kHz is no big trick. You filter off the midrange bump (if any) and you're all done. Instant 700Hz crossover. But the Khorn/Belle/LaScala are another story altogether - getting the bass and mid horns on speaking terms is not a simple matter. It can be brute-forced by going with a *big* Altec theatre horn, or an even more gigantic Edgar salad-bowl horn. These are all big, room-dominating horns, looming over the top of a Khorn. Al K has a more elegant approach with an ultra-steep crossover, maybe even an elliptic filter, in order to keep the horns well-behaved on the edge of their usable range. But no free lunch there either; elliptic filters have substantial phase distortion, but then again, the several-millisecond path-length difference between bass and mid horn probably renders phase distortion moot. The waters get really deep with an all-horn system, mostly around the bass-mid transition. The Chorus lets me take little baby steps with horn systems, letting me find out what horns like to do and what they don't like to do. I think my next experiments will be with the Beyma tweeters - the mid-tweeter crossover is a lot simpler.

Sorry - Beyma CP22 and CP25, not the numbers posted above. The two look different, and each has its appeal. The CP22 has noticeably lower distortion than the CP25, and the horn profile at first glance seems smoother considering the throat-to-horn-mouth distance. By contrast, the CP25 has (much) wider dispersion, and more extended mid response, simplifying the crossover design. Anyone who's been using the Beyma tweeters is welcome to chime in with their experience.

The Beyma CF22 and CF25 look interesting as replacements for the stock tweeter, if the Beyma curves at US Speaker are to be believed. Efficiency is higher, but with the arrival of the new autoformers, correcting for that is no problem. There is some evidence of smoothing on these curves, but not as heavy as Fostex, where the factory curves look pretty different than the MLSSA data. (Warning! The Fostex PDF is a 10MB download!) Very likely that Gary will use the Altec 311's on his Chorus', and I will keep the stock mid and try the Beyma tweeters in small sub-enclosures sitting on top of my Chorus I's. That way I can time-align the speaker without a lot of trouble, and get better treble while I'm at it.

I'd like to throw in my two cents worth in favor of 96/24 DAD's. They play on any DVD player, including the fancy SACD/DVD machines, and most recording systems these days, even for entry-level DAW systems, support the 96/24 format. What's annoying about 192/24 DVD-A is that the medium defaults to a miserable 384kps Dolby Digital if the player isn't a late-model DVD-A player, and worse, consumer-grade DVD-A players leave a lot to be desired sonically unless you pay kooky prices for Krell or suchlike. As we know, the matter of transporting high-rez digital out of a transport to a high-rez external DAC is mired in Hollywood copy-protection schemes and legal shenanigans. SACD/DSD is hostile to home recordists, with almost nothing available for DAW systems. The front-end stuff is very thin on the ground compared to 96/24 equipment, which seems to be everywhere. Which leaves 96/24 as a format that plays on ALL DVD players - from the cheapest COSTCO special to super-high-end - and is well supported on the recording end. My question, though, is 96/24 DAD mastering software hard to get or use?

"Lynn - you mentioned that they SRC'ed from 24/192 down to the other formats? To me, this is already a problem. SRC is not transparent on any system short of the absolute top-shelf stuff (and no, Apogee is not top-shelf)" In all honesty, I don't know what Meitner et all were using for SRC - although K. Johnson's gear would be a safe guess. I'm not a recording guru, so I don't know the status of Meitner/KJohnson stuff in the pro world. Might be good, might be crummy, dunno. I have to confess I'm mostly an analog guy, but a lot of that is sour grapes from the early Eighties, when the recording industry made the rapid switch from analog to bare-bones 44.1/16 without dithering and the worst A/D converters imaginable. I still think of the Eighties as a lost decade, when recorded sound quality actually went backwards for the first time since the invention of the phonograph. It took PCM a long long time to get up to 15IPS mastertape quality, something like 15 to 20 years realistically. I'm old enough to remember the conversion from mono to stereo - that was a LOT more effort all around, and took less than 5 years, and involved media as diverse as pre-recorded tapes, movie-sound, LP's, and FM radio! So the "hard-to-do" excuse doesn't cut a lot of ice with me. Stereo sound was a MUCH more difficult and costly conversion process, and happened three times faster. Here we are, a full twenty years after the introduction of the CD, and high-rez finally equals the quality of an all-analog mastertape made in 1975. 20 years of effort - for what? So people can walk around with MP3 players that sound just a little better than 1970's cassettes?

Hi folks, thanks for the great discussion!!! Lots of good input from folks who do this for a living, people who ought to know. The 2001 VSAC demo was a long way from a scientific double-blind test, but I've heard Keith Johnson's classical recordings at the CES before and this demo considerably exceeded what I heard then. The recording and playback gear was pro-grade stuff from Ed Meitner and Keith Johnson, actual data sources were removable hard disks, and Johnson was playing recordings of Eastern European symphony orchestras that he made for reference purposes (although some are commercially available as downmixed HDCD recordings). Although the sound was very good - certainly a lot better than the dreadful sound you usually hear at the CES, or worse, the S'Pile shows - it was not the best sound I've ever heard. That was visiting the BBC in 1975 and hearing a discrete quadraphonic recording of Beethoven's 9th Symphony, and listening to a first-generation mastertape made on a customized Studer running at 30IPS with no Dolby processing. Speakers were professional BBC monitors with internal bi-amplification and the microphone was a Clarec Soundfield prototype. I have never before or since heard any recording with the startling sense of realism and audience presence that one had - at one point, the audience got excited and started clapping along with the music, and you could hear every handclap with crystal clarity, in every direction (this was 360 degree quadraphonic mastertape), along with the 100-strong chorus going full shout - all with NO distortion whatever. Usually a dense piece like the 9th is just a roar of hash and distortion at the climaxes - not this recording. It certainly would be fantastic if the BBC ever released this on multichannel SACR or DVD-A - it would be hands down the best recording of the 9th ever made. To descend from the sublime to the geekiest, there's bunch of MLSSA graphs and stuff over at the Odds and Mods forum.

Here's an Energy-Time-Curve the TEF guys like to use, although I don't use it very much myself. It basically shows the overall decay in energy over time, but in dB vs time. It shows nothing of which frequencies are dominating the energy storage - which is why I don't use it - but it does provide a rough figure-of-merit showing how quickly a speaker quiets down after it gets a transient. The ideal speaker would shut up instantly, but we have to live with real-world driver with physical mass and many many resonances. By the way, most audiophiles misunderstand the term "transient response". They think it's the risetime, but that's trivially a function of HF extension - in other words, the higher the HF extension, the shorter the rise time. Morever, the ear mostly perceives the rising edge as nothing more than a click. The real differences in speaker drivers, and audibility, is the decay time, where the ear/brain has something to go on, and is looking for echoes and resonances. It would be nice if all you heard was musical resonances and echoes from the performaning space, but real speakers interfere with their own set of reflections and resonances that have nothing to do with musical values. Chasing out these resonances is the hard part of speaker design, since they are inevitable with physical drivers, horn-mouth reflections, and cabinet diffraction.

And here's the CSD of the modified Chorus. I've picked out the dominant resonance, which seems to be around 6kHz. Note this high-Q resonance doesn't appear at this frequency on the freq response graphs shown earlier - this is a good example how freq resp graphs are misleading when tuning notch filters. You want to chase out the high-Q resonance you see on the CSD, not necessarily the bumps on the freq response curves. The high-Q resonances, since they are so narrow, are not so much excited by musical harmonics as they are by musical transients, which have a broad spectral excitation. At these frequencies, resonances lend a metallic coloration, due to the bell-like high Q and fairly long decay time. A long 2.5mSec decay may not seem like much, but remember, that's a decay tail that's nearly 3 feet long!

So far you guys have been getting off light, now I'm going to throw you the serious data. The Cumulative Spectral Decay, 3D graph, or Waterfall curve shows time, frequency, and resonant structures all in one graph. The rear-most curve is the unsmoothed freq response - in fact, most MLS systems provide no means for smoothing this data, so if you're suspicious of the mfg's freq curves, take a close look at the Waterfall curves and look at the real, unsmoothed freq response. D.E.L. Shorter of the BBC first wrote of "delayed resonances" in Wireless World in the mid-Fifties, and this referred not to major resonances that show in freq resp graphs, but subtler resonances that continue to ring on after a musical transient. Kind of like a quiet off-key bell that plays alongside the music - and all drivers do this, the only problem back then was measuring it. Shorter used an exotic 10-cycle-on, 10-cycle-off chopper circuit that triggered a scope display when the sine wave cut off, allowing him to visualize narrowband stored resonances as he gradually swept an oscillator over the working passband of a driver. Unfortunately transatlantic prejudices and the old Not-Invented-Here syndrome kept this technique out of the USA for more than twenty years. It wasn't until the Crown TEF system, designed CalTech's Richard Heyser in the early Seventies, that allowed quick visualization of the stored resonances. The advent of PC-based MLS systems in the late Eighties brought powerful analytic tools to the PC owner for the first time when the Doug Rife MLSSA system hit the market. These days, you can do this kind of analysis with a sound card, although a high sample rate (at least 96kHz) and a built-in lowpass filter set to 25kHz or higher is still desirable. The measurements you see here, for example, have a sample rate of 117kHz and a Butterworth lowpass set to 25kHz (the MLSSA system has tunable lowpass filter, allowing choice of frequencies and a selection of Chebychev, Butterworth, or Bessel filter shapes).

And here's the Gary Dahl-modified Chorus with the Fostex tweeter. No difference in the two-spike time response, but you can see the mid driver is attenuated by about 2dB relative to the stock Chorus, and the Fostex has a little tidier reflection performance (fewer ripples in the time domain). The reversal of phase for the mid horn is also visible compared to the stock speaker.

So much for the easy stuff, guys. Now I'm going to show the raw time data from which MLSSA calculates the freq response. This is the stock Chorus II in the time domain. The twin-peak structure reveals that it has mid and tweeter horns with dissimilar path lengths to the microphone (1mSec equals 14" distance). It is standard design practice to trim the phase response of the crossover to smooth out the crossover region, but the 1mSec discrepancy in the time domain remains and is visible on MLS, FFT, or TDS systems. There are two "fixes" if the design thinks close time-alignment is desirable: either move the HF tweeter towards the back of the cabinet, sitting just behind the mid voice coil, or use a digital crossover and slide things around in the digital domain. Most speaker designers (but not all) believe time distortion of 1mSec or less is not audible. Delays of 3mSec, though, are more controversial, with the argument going back to the original Shearer horn for MGM, and the resulting Altec 2-way theatre systems.

Here's a different view, with 1/10th octave smoothing and the FFT window opened up to 20mSec. The notch around 200Hz is caused by comb filtering from the floor bounce (all speakers do this), and the smoother appearance of the HF is an illusion created by the smoothing algorithm. If smoothing is not used, the traces appear very rough due to multiple room reflections falling in the 6-20mSec window. Those fortunate enough to make free-field measurement in an anechoic chamber, or outdoors hanging from a crane, can use FFT windows as large as 20mSec or more, removing the need for smoothing. The use of smoothing is controversial in speaker-design circles. Some feel it makes graphs easier to read by directing the eye to broad trends, and others feel it hides fine detail, making for prettier graphs the marketers like. I'm in the second group, but I'm used to the wrinkles that real-world drivers have. It should also be noted that 1/10th octave smoothing is MUCH LESS than many driver mfg's like to use ... 1/3 octave smoothing (much heavier) is pretty typical when measuring rough PA drivers. By comparing these two graphs in your browser, you'll get an idea of the appearance of smoothing, so you'll know it when you see it in mfg's data sheets.

Here's the first image, showing the before-and-after of Gary Dahl's Chorus II's. The green trace shows the modified system, the orange trace the original. This is a 6mSec FFT that excludes room reflections, and should not be considered accurate below 300Hz (due to window length). It is, however, an unsmoothed graph, so the full depth of the HF nulls are apparent.