I think you missed my point. Back EMF from the driver can allow the amplifier to control the voice coil movement, however the middle of the dome basically flops about doing it's own thing at high frequencies as it's only very loosely coupled to the edge because of it's own less than infinite stiffness. Thus any distortion or resonances that occur due to the middle of the dome bending and moving in non-piston ways are not reflected back to the amplifier via back EMF.
References please ? How exactly does a standing wave form along the ribbon when the entire ribbon surface is driven together in phase ? This is exactly what does not happen on a ribbon. What you describe is the case with a soft dome however, where all kinds of complex standing waves patterns are occurring at high frequencies.
Highly debatable, depending on the model.
Most modern ribbon tweeters are wave-guide loaded, while most dome tweeters are not. This increases the sensitivity and maximum SPL considerably. They also have exceptionally strong magnets, so sensitivity is no longer an issue today.
My AC G2 are 96dB/W/M sensitivity - a lot more than most modern domes which are around 89-92dB.
Sure, their true RMS continuous power rating is around 4 watts, but with 96dB sensitivity that's 102dB SPL continuous, peak output on transients can be much higher. That's for a ribbon that's only 50mm x 8mm, larger ones have both higher power rating and higher sensitivity again, (some over 100dB/W/M) so dramatically higher SPL potential.
Dome tweeters are not much higher in true RMS power rating - all tweeters (both dome and ribbon) are not rated by their true raw RMS power rating, they are always rated based on program material with a high pass filter in place - for example if they're rated at 50w that is for 50w of "average" program material or pink noise with a recommended high pass filter in place (say 3Khz 12db/oct) - the actual power reaching the tweeter will be about 1/10th of that.
The true power rating of most non ferro fluid 25mm domes is around 4-7 watts, with ferro-fluid models maybe reaching 10 watts. Given that they're often a lot less sensitive and need more power to produce the same SPL, "limited power handling and SPL" of a good ribbon is not an issue like you suggest it is, let alone with a bigger ribbon that is much more capable than a dome in SPL output.
Whether the ribbon is narrow or not has nothing to do with vertical dispersion, it's controlled only by the vertical length. Yes, long ribbons (100mm) are somewhat directional vertically, but I really don't think it's an issue with a 50mm ribbon in a design where its near ear height.
Vertical beamwidth at 20Khz is about 40 degrees, and about 80 degrees at 10Khz, I don't find that a problem at all. One beneficial effect of reduced dispersion vertically is far less treble reflection off the floor and ceiling, leading to a lot better imaging at far listening distances.
Certainly in 8 years of using them I've never felt "oh I wish they had wider vertical dispersion". The roll-off in the top octave when listening while standing is very smooth and subtle.
Yes, but when the ribbon is only 8mm wide compared to a 25mm dome, there is far less narrowing of dispersion with increasing frequency than a dome. The directivity control is achieved with a wave-guide instead. This is why a wave-guide loaded ribbon can achieve an almost constant 90 degree horizontal dispersion from 2Khz right up to 20Khz - the ribbon element itself is far less directional horizontally at high frequencies than a dome, with the wave-guide then adding in a constant directivity control.
This constant directivity in the horizontal plane means that the early side-wall reflections are spectrally flat in response, something which is not the case with a dome tweeter where the dispersion is far greater than 90 degrees below about 5Khz but less than 90 degrees above 10Khz. The early side wall reflection is also lower in amplitude, especially if the angle of incidence between the speaker/wall/listener is greater than 45 degrees from the speaker.
No, no free lunches, with a ribbon you do have to cross it over higher and steeper (I cross mine at 4Khz 18dB/oct) and you have to choose between a short ribbon (~50mm) which has good vertical dispersion and similar SPL capability to a dome, or a long ribbon which is more directional in the vertical plane, but has MUCH greater SPL and dynamic capability than a dome. Tough choice
Other than that there aren't really any drawbacks. One thing is for sure, I can't stand sibilance and if a ribbon tweeter is one of the few driver types that can eliminate it, that alone makes it the best choice for me. The fact that they also sound incredibly neutral, eerily realistic, and are completely un-fatiguing to listen to is a nice bonus.
Acoustical is what matters; how the acoustical result is achieved is irrelevant. Direct-radiator (no horn) drivers operate in a constant-acceleration regime, and acceleration is the 2nd derivative of motion. Thus, there is a 12 dB/oct slope in the bass direction for an ideal direct-radiator with flat response. This slope terminates at Fs, and the excursion response stops rising and goes flat below Fs.
Adding a crossover adds a tilt below the crossover frequency, which decreases the "area under the curve" compared to no crossover. As you can see, the real function of the highpass filter for the tweeter is to decrease this area under the curve, protecting the tweeter from outright damage as well as lowering IM distortion. And all tweeters have IM distortion. Distortionless drivers do not exist. Some are better than others, but don't expect to find a 100x reduction. The difference between the best and the worst is about 10X.
Unlike IM distortion in an amplifier, the real-world distortion of a driver with a musical stimulus (dense spectra) is caused by low frequencies crossmodulating with higher frequencies. Since the dominant form of driver distortion is the result of excursion, you can see how relatively low levels of out-of-band energy (not directly audible as low frequencies) can seriously affect the spectral cleanliness of the higher frequencies (the in-band sounds).
For example, say the crossover isn't doing it's job at 300 Hz, the frequency that typically is strongest on many recordings. This means there will be a picket fence of +/- 300 Hz sidebands for musical tones in the passband of the tweeter, typically 3 kHz on up. With simple material, this may not be audible, but for dense material, where many singers are very slightly off-pitch and creating a very dense spectrum, the additional sum-and-difference terms can add a layer of "hash" that sounds quite unnatural and "electronic".
At moderate levels, there is loss of resolution - "veiling" in the language of reviewers. Usually, the problem is much worse, resulting in obvious artifacts, a sort of harsh buzzsaw sound, and if really severe, outright breakup with tearing and roaring sounds. I commonly hear artifacts like this in hifi demos at shows, and have been asked to leave the room and take my recordings with me.
Many so-called "bad" recordings have a combination of dense spectra, a "hot" tilted-up balance, and are recorded at a high level. These can slew a phono preamp or the analog stages of a CD player, making things sound much, much worse. On the other hand, if the analog electronics are fast enough, and the tweeter has a well-designed crossover, the same "unplayable" recordings can sound absolutely beautiful - maybe a little bright, but vivid and thrilling. Put it on another system, and the sound becomes grossly distorted and unlistenable.
How fast? Well, I've measured what was coming out of a Burr-Brown PCM63 ladder DAC before it hit the lowpass filter (20 kHz stimulus), and I saw a comb spectra that was flat to 20 MHz. The comb finally disappeared into the 80 dB noise floor of the spectrum analyzer at 50 MHz. The comb itself was composed of a picket fence of 20 kHz sidebands, of course. A back-of-the-envelope calculation revealed that the analog electronics for the first stage (the lowpass filter) needed to be about 1000 V/uSec to avoid slewing that comb spectra. Guess what opamps are commonly found in "audiophile" CD players? The 30-year-old 5532/5534, which has a slew rate of 13 V/uSec.
Replace the first-stage opamps with ultrafast video amps, and you know what happens? Those so-called "bad" recording now sound just fine, no HF congestion at all. Same story for phono preamps. Increase the slew rate and headroom, and all those annoying ticks, pops, and end-of-side distortion suddenly becomes much less annoying. Preamp slewing and overload is much more common than people think - even the best cartridges spend a fair percentage of the time mistracking (look at scope pix), and every time the stylus loses contact with the groove wall, that generates a very fast transient that can slew a preamp. Slewing stretches out the loss-of-groove-wall transient and makes it a lot more audible.
Last edited:
Thanks Lynn, an excellent explanation and gives me a lot to ponder
One thing that immediately springs to mind is an asymmetric filter on the tweeter Something that provides the desired acoustic rolloff down to the point where the tweeter is no longer significantly contributing in frequency and then sharply cuts off below that. Not so easy with passive I would imagine though, and could be tricky phase wise...
Tony.
One thing that immediately springs to mind is an asymmetric filter on the tweeter Something that provides the desired acoustic rolloff down to the point where the tweeter is no longer significantly contributing in frequency and then sharply cuts off below that. Not so easy with passive I would imagine though, and could be tricky phase wise...
Tony.
Well, yes, that's the other half of the problem. You want the midbass and tweeter to phase-track each other within the crossover region, to avoid obvious phasey and incoherent-sounding artifacts. I aim for 10-degree or better phase tracking through the crossover region. If you have an inphase style of crossover (acoustical 2nd or 4th-order), temporarily inverting the phase of the tweeter should reveal a nice deep null at the crossover point, which is an indication of good phase tracking. If your crossover has a 90-degree phase angle between the drivers (acoustical 1st or 3rd-order), well, you're on your own.
By the way, now you know my reaction when I hear that a big-name speaker designer has "hired out" the crossover design to a third party. Uh - crossover design IS the speaker design. "Hiring it out" means the so-called designer is really more of a marketer than anything else.
By the way, now you know my reaction when I hear that a big-name speaker designer has "hired out" the crossover design to a third party. Uh - crossover design IS the speaker design. "Hiring it out" means the so-called designer is really more of a marketer than anything else.
Last edited:
About a year or two ago, Parts Express had an offer for 200 3/8" mylar (polypropylene) dome tweeters for $10 a box. These are packaged as small cylinders with leads attached. I bought two boxes of them. I also bought close to 200 Sanyo 3/8" mylar domes mounted in 6 x 9 adaptor rings with 3.3 mfd caps for 28 cents each. These were intended to upgrade automobile 6 x 9 speakers in high end GM cars.
I find these tweeters extremely useful. One advantage of an array is that you can direct the drivers to produce whatever propagational directional characteristics you want.
Another advantage is that they draw so little current that you can create series and parallel arrangements of them with combined impedances of as low as one ohm and they don't seem to bother any amplifiers including HT types at all. I've got five per channel wired in parallel with Empire 9000M connected to a Pioneer HT receiver and there is no problem. Sensitivity usually ranges at around 90 to 91 db and I prefer the 8 ohm versions but they also come 6 and 4 ohms. These tweeters seem to me to be similar to Audax TW010F1. I've bought those branded as Audax, Radio Shack, and Dayton. Parts express joked that there are 28.5 billion of them on the planet.
There are other variants of these tweeters. I avoid the titanium domes for the reasons stated elsewhere and there's also a cloth dome version. These tweeters are usually rated to handle around 15 to 30 watts and have an FS of around 2900 to 3000 hz. I never cross them over at less than 6khz and sometimes up to 9 khz, usually with a single capacitor for a 6db per octave slope. I've also used a 1 mfd cap across the voice coil to roll off the high end where they fire directly. Cheap, small, versatile, easy to use I've been re-engineering all of my speaker systems with them during the last 22 years. Experimentation usually yields excellent improvement at very low cost for speakers others would normally consider hopeless.
Hear hear. I couldn't have agreed more with everything you said
I think it's grossly underestimated by many people just how serious the problem of intermodulation distortion from out of band signals is with tweeters.
Any time I see a design with a dome tweeter crossed over with a 1st order filter (or even 2nd order for that matter) a part of me shudders to think how much out of band excursion related distortion there is. Then there is the trend towards crossing tweeters over very low - 2Khz or even less. Again, I shudder. Some designs even use the natural roll off of the tweeter below resonance as part of the roll-off slope and design around that. More shuddering.
Serious intermodulation distortion and power handling problems are introduced all in the name of what - slightly better polar response ? No thanks.
It's exactly as you say - the sound gets very congested and unpleasant at high volumes because of all that IM distortion from dense spectra music.
A couple of other points I would add to yours related to this is the frequent lack of any RLC resonance compensator for the fundamental resonance of a tweeter in a passive design which can cause a LOT of additional excursion at the tweeters self resonance - heard as a raspyness whenever there is a strong midrange component to the music.
A second common problem is a passive high pass filter for the tweeter where the shunt coil doesn't have a low series resistance - the filter might start at 12 or 18dB/oct when first rolling off, but will revert to ~6dB/oct well before reaching the lower midrange where as you mention most of the power in the music is.
By using a coil with higher resistance your measured frequency response will look fine, (since the slope wont deviate until well below the passband) but the tweeter may be receiving 10-20 dB more power at low midrange frequencies than it should be compared to an ideal roll-off slope.
So it's VERY important for the shunt coil in a tweeters crossover to have very low series resistance, <0.1 ohm if possible. (Or just use an active crossover...)
Using ribbon tweeters that by their nature have to be cut off higher and steeper (at least the smaller ones) has really taught me the virtue of crossing tweeters over higher and steeper, and putting more effort into finding a midrange driver that can work well higher in frequency rather than a tweeter that can go lower.
Even a small increase in crossover frequency can cut both power handling and excursion requirements of the tweeter dramatically (without putting any significant additional strain on the midrange driver) as there is so much more power in the midrange than the treble in music.
Years ago I used an audio processing application to examine a representative set of music - applying software based high pass filters of various slopes and cutoff frequencies, and measuring the 1 second averaged RMS power that would be delivered to the tweeter.
I compared 3Khz 12dB/oct, 3Khz 18dB/oct, 4Khz 12dB/oct, and 4Khz 18dB/oct.
The result was that by simply increasing the crossover point from 3Khz to 4Khz, for a given slope, the power handling requirements of the tweeter are reduced by 3dB, eg halved. Going from 12 to 18dB/oct at the same frequency reduced it by a further 1.5dB, so in total going from 3Khz 12dB/oct to 4Khz 18dB/oct reduced the power handling requirements by 4.5dB, which is not to be sneezed at.
Of course when you look at the excursion requirements and IM distortion, the filter slope becomes more important than the crossover frequency for the excursion vs frequency reasons you outlined.
With a ribbon crossed at 4Khz 18dB/oct and midrange driver that can truly overlap it and work cleanly at high frequencies I find even at very high volume levels there is no congestion or raspyness on dense spectra music that you might find on a dome tweeter crossed over much lower and at a shallower slope.
I'm more than willing to sacrifice a little bit of off axis power response to achieve this.
"The difference between the best and the worst is about 10X."
So would you say that 10 or even 20 or 30 of my mylar domes will do as well as the TOTL from Scanspeak and Morel? Look at the cost difference, a few dollars per channel for mine compared to hundreds for theirs. Well if true that is good news. BTW, I think my arrangement works quite well and when carefully equalized solves the problem of sibilance and many others.
"On the other hand, if the analog electronics are fast enough, and the tweeter has a well-designed crossover, the same "unplayable" recordings can sound absolutely beautiful - maybe a little bright, but vivid and thrilling."
I haven't been thrilled about much of anything lately. Give me a few recommendations that I will find thrilling!
I recommend the Mercury Living Presence transfers to CD, or if you have a good phonograph, the originals. These records, unlike the forgiving RCA "Living Stereo" recordings, can be nearly unplayable on many high-end systems - just an appalling screech instead of massed violins, and especially offensive to anyone that likes the sound of real, unamplified music. But ... if the slew rate of the electronics are high enough, and/or free of Class AB artifacts, and the tweeter crossover is correctly designed, it's a thrill a minute, with astonishing performances that just sound a little bright and up-close.
The reason I mentioned the spread in IM distortion between the best and worst tweeters is that seemingly minor changes in the crossover overshadow differences in the tweeters themselves. When I designed the Ariel in 1993, I chose a Scan-Speak D9000 tweeter that had a very long (for a tweeter) 1mm excursion, then selected an acoustical LR4 crossover at 3.8 kHz. Many people have told me that the Ariel would be "better" if I had a crossover like the rest of the industry - less slope, lower frequency, much prettier polar pattern. To which I say, "No it wouldn't - not if you care how it actually sounds". What also made such a high crossover possible is the very smooth rolloff of the Vifa 5.5" driver - to this day, one of the smoothest-rolloff drivers I know of. Most rigid-cone modern drivers are much worse, and require a more aggressive crossover (with notch filters) to suppress out-of-band resonances.
The two points raised by DBMandrake are very, very important. Control of the Fs region is extremely important - not just for excursion control, but also good phase relationship between the midbass and tweeter. As Laurie Fincham of KEF showed me back in 1975, you must synthesize an acoustical transfer function that goes right through the Fs region - it's the only way to make sure the inter-driver phase angle is well controlled (which prevents sudden changes in vertical polar pattern, which are quite audible), as well as keeping tweeter IM distortion to a minimum. Laurie had to tame very tempermental KEF T27 tweeter, which had almost non-existent midrange power-handling, so he knew what he talking about. (At Audionics, I was assigned a speaker with the T27 tweeter, and I would never use it again - very limited power-handling, and HF response that was just OK at best. One of the more difficult and finicky tweeters I've worked with. Modern tweeters are much, much better.)
The DC resistance of the shunt inductor for the highpass filter is a subtle problem that trips up a lot of people. (Big shout-out to George Short of North Creek Music Systems for pointing this out to me.) It's counter-intuitive to use a massive low-DCR coil in a tweeter circuit, but yes, the value of the DCR sets the transition point where an electrical 2nd or 3rd-order filter transitions back to an ineffective 1st-order filter. You really don't want 100 to 300 Hz sneaking into the tweeter - even if you listen to the tweeter by itself (a powerful debugging technique to reveal crossover or tweeter problems), you'll never hear 100 to 300 Hz being radiated by the tweeter. It just can't do that, the acoustical output is too far down. But ... the IM distortion most certainly will be there, audible as an intermittent loss of transparency with certain combinations of program material. (This, by the way, is a very common fault in even the most expensive audiophile speakers. It's another reason they use simple program material for hifi show demos - to conceal the design faults of the HF system, which is an industry-wide problem.)
The reason I mentioned the spread in IM distortion between the best and worst tweeters is that seemingly minor changes in the crossover overshadow differences in the tweeters themselves. When I designed the Ariel in 1993, I chose a Scan-Speak D9000 tweeter that had a very long (for a tweeter) 1mm excursion, then selected an acoustical LR4 crossover at 3.8 kHz. Many people have told me that the Ariel would be "better" if I had a crossover like the rest of the industry - less slope, lower frequency, much prettier polar pattern. To which I say, "No it wouldn't - not if you care how it actually sounds". What also made such a high crossover possible is the very smooth rolloff of the Vifa 5.5" driver - to this day, one of the smoothest-rolloff drivers I know of. Most rigid-cone modern drivers are much worse, and require a more aggressive crossover (with notch filters) to suppress out-of-band resonances.
The two points raised by DBMandrake are very, very important. Control of the Fs region is extremely important - not just for excursion control, but also good phase relationship between the midbass and tweeter. As Laurie Fincham of KEF showed me back in 1975, you must synthesize an acoustical transfer function that goes right through the Fs region - it's the only way to make sure the inter-driver phase angle is well controlled (which prevents sudden changes in vertical polar pattern, which are quite audible), as well as keeping tweeter IM distortion to a minimum. Laurie had to tame very tempermental KEF T27 tweeter, which had almost non-existent midrange power-handling, so he knew what he talking about. (At Audionics, I was assigned a speaker with the T27 tweeter, and I would never use it again - very limited power-handling, and HF response that was just OK at best. One of the more difficult and finicky tweeters I've worked with. Modern tweeters are much, much better.)
The DC resistance of the shunt inductor for the highpass filter is a subtle problem that trips up a lot of people. (Big shout-out to George Short of North Creek Music Systems for pointing this out to me.) It's counter-intuitive to use a massive low-DCR coil in a tweeter circuit, but yes, the value of the DCR sets the transition point where an electrical 2nd or 3rd-order filter transitions back to an ineffective 1st-order filter. You really don't want 100 to 300 Hz sneaking into the tweeter - even if you listen to the tweeter by itself (a powerful debugging technique to reveal crossover or tweeter problems), you'll never hear 100 to 300 Hz being radiated by the tweeter. It just can't do that, the acoustical output is too far down. But ... the IM distortion most certainly will be there, audible as an intermittent loss of transparency with certain combinations of program material. (This, by the way, is a very common fault in even the most expensive audiophile speakers. It's another reason they use simple program material for hifi show demos - to conceal the design faults of the HF system, which is an industry-wide problem.)
Last edited:
Sure if you assume "in phase" back EMF - I am suggesting it is never "in phase" therefore the result is "distortion". No matter how small in level one wishes to presuppose this might be...
Perhaps so. But ur assuming a dome that has flop... it may not be a dome.
There are other modes in a dynamic driver that cause distortion. And, while this is certainly a source of distortion that needs to be considered, it's one of many that are problematic.
You mean IF the entire surface is driven in phase?
To begin with the very top of the ribbon is usually NOT driven equally as is the center portion. The maximum xmax is in the center. The two ends are restrained and that portion coming from those points acts more like a compliance than a linear portion of the ribbon. I think if you measure the ribbon you will find that most of the highs measure best and greatest in amplitude coming off the dead center of the ribbon... at least that has been my experience
I suggest you look at the Decca Ribbon. They required silicone grease to be placed approximately at the 1/3 points to maintain ribbon stability due to the various modes and standing waves that naturally occur.
I think you will find that the "waveguides" are present so that the ribbon can be used lower in frequency, and also to control the dispersion when used in PA/SR apps... the waveguides have little or no effect higher in frequency, and often cause a less flat response...
Ok. I'm not knocking ur tweeters - but compare them to the rating of the SA (now defunct) and similar modern tweeters - albeit printed foil on kapton or similar substrate - which do SPL levels around 128dB/SPL, have power ratings around 40 watts, and in the case (as previously mentioned) of the SA, measured THD at 128dB of <1%...
Basically I agree - but I'm not advocating dome tweeters as any sort of definitive way to go. However their polar response in vertical and horizontal tends to be better in some regards than do ribbons, planars and similar drivers. So there is a trade off...
How about an ESL?
I disagree.
Take ur ribbon outside, where the room reflections will not color your opinion and have a listen. Let me know what you hear?
If your ribbon, which I am not familiar with, and don't have a link to see (post a link?) are basically a "square format" meaning a short squat ribbon, then sure, you can get that dispersion (maybe) but at the cost of power handling and SPL.
As you go toward the top end of the max SPL level, ur distortion is rising a whole lot.
My main horns are 108dB/1w/1M... max power is rated at 40w continuous...
so at 90dB SPL, work the numbers... I prefer big headroom with low distortion.
I have not seen that to be the case, would like to see a ribbon/waveguide that does this. And it won't do that vertically - as I said the purpose of the waveguide is to control + boost the lower freqs, not the higher freqs...
I think we're getting a bit far afield now... but the issue you raised is complex and has little to do with "sibilance".
No, no free lunches, with a ribbon you do have to cross it over higher and steeper (I cross mine at 4Khz 18dB/oct) and you have to choose between a short ribbon (~50mm) which has good vertical dispersion and similar SPL capability to a dome, or a long ribbon which is more directional in the vertical plane, but has MUCH greater SPL and dynamic capability than a dome. Tough choice
Ummm... what do you use to meet the ribbon at 4kHz?? That's a problem I try to avoid.
_-_-bear
Lynn, ur saying the I/V converter op amp??
The usual path is an I/V conversion, then a LP filter, then a gain stage, then a buffer...
I use a gyrator circuit for the LP filter - and the opamp selection there altered the sound, as did the gain stage. The buffer I use is a discrete jfet/mosfet circuit...
I have tried some high speed video opamps and they were awful.
I'm wondering if you arrived at a specific opamp or feel that all fast video opamps do the trick?
_-_-bear
Lynn, Thanks for the tip. I'm sure I've got quite a number of both original vinyls and of cd transfers of Mercury Living Presence recordings. I think these were made with just two or three mikes. MCA I think is one of their reissue lables. Westminster was also one of their early record labels. I'm sure I've got a vinyl of the Scotch Symphony and of the Pastoral. I'll keep an eye open for them, my collection is not well organized. OK the vinyls are completely disorganized....all 3000 of them
I think this is why they're euphemistically called "waveguides". If they coupled well across the driver's entire useable bandwidth, they'd be proper horns.
For the record there is a definition of the distinction between a waveguide and a horn - I think it is discussed extensively in the Geddes thread? I can't quite get the swiss cheese excuse for my brain to pull it out...
_-_-bear
PS. Lynn, I am surprised by your report on the Mercury Living Presence recordings, were they not done onto optical film?? I have yet to hear them to good advantage. I was given a re-issue - I don't recall if it is vinyl or CD - and may have to revisit it based on your recommendation.
_-_-bear
PS. Lynn, I am surprised by your report on the Mercury Living Presence recordings, were they not done onto optical film?? I have yet to hear them to good advantage. I was given a re-issue - I don't recall if it is vinyl or CD - and may have to revisit it based on your recommendation.
Matt Kamna and I did these tests in the mid-Nineties (Matt and I are old Tektronix guys, and we collaborated on a number of projects when I lived in Portland). Anyway, Matt has a rebuilt $40,000 HP RF spectrum analyzer that was good out to 100 MHz and an on-screen noise floor around 80 dB. Just the ticket for seeing what's going on inside a DAC. I was modifying a Monarchy M-33 DAC that I owned at the time, and I wanted better performance from the analog section (which was actually pretty good in stock form).
First step, we twiddled around with passive I/V conversion, and discovered out that the Burr-Brown PCM-63 is just fine with any resistor value of 100~120 ohms or less. Any higher, and the internal diodes inside the PCM-63 start to turn on and distortion goes up. Any lower, and distortion is at or below factory-spec levels (we had other spectrum analyzers that could measure below the specified distortion of the PCM-63). So aside from noise considerations, the PCM-63 is perfectly happy with passive I/V conversion.
Just to double-check what the requirements for active I/V conversion and following active low-pass filtering might be, we directly measured what was coming out of the 100 ohm resistor - the RF analyzer was plenty sensitive and had an internal 50-ohm input on the BNC connector, so no problems there. We threw a 20 kHz full-modulation test tone at the DAC, and sure enough, the scope shows exactly what you'd expect - nice square waves with very fast leading and falling edges. In fact, they were so fast they were kind of hard to measure, so the RF spectrum analyzer was the most useful tool to assess what was really coming out of the DAC. After all, how could you design an active I/V stage and associated active lowpass if you had no idea of the real-world stimulus? (Tek thinking here - measure first, guess later, not the other way around.)
Well, we were shocked to see a full-power comb spectra going out to 20 MHz. I had thought maybe it might attenuate a little beyond the audio range, but no such luck. It was full power right through the shortwave band - add a RF transmitter, and it would be a good enough noise source to get us in trouble with the FCC. The comb didn't extinguish into the noise floor until 50 MHz, probably due to assorted strays in the test layout.
Well, after discovering that, it put the requirements of an active I/V stage in a different light. You need an extremely fast slew rate, very low distortion, and very low noise = and the low distortion has to be maintained out to 20 MHz!!!
Almost no off-the-shelf solutions exist for these requirements. The malign thing about slewing is that it doesn't appear that strongly in audio-frequency THD or spectral analysis measurements, because the interval of slewing is so short compared to the overall cycle of the waveform. That doesn't change the fact that slewing represents a brief interval of complete loss of information, or stated otherwise, 100% distortion. There is no possible way the data lost in the slewing interval can be reconstructed, even in principle, unlike conventional forms of distortion. The fact is it is very brief doesn't change the fact that information is being irrecoverably destroyed, in the same way mistracking or hard clipping represent intervals of information destruction.
A lot of designers mistakenly assume that slewing and low-passing are comparable; they are not. Low-passing should be distortionless, and is an essential requirement for Nyquist reconstruction of analog data. Slewing, although superficially similar, dramatically raises HF distortion, and in a very complex program-dependent way. It undermines the accuracy of Nyquist reconstruction, which assumes infinitely fast and accurate lowpass filtering. (In other words, if the active lowpass filter slews, Nyquist reconstruction is violated, and additional distortion is introduced.)
The requirement for very low distortion in the MHz range is very difficult to accomplish, and a powerful argument for some form of passive pre-filtering before any active circuitry is encountered. Since audio circuits typically have rapidly rising distortion above 100 kHz, that's probably where the first pole should go - certainly, not much higher than 200 kHz.
Matt and I tried various digital-video buffers, and some had good audio performance, and others didn't. They were the only off-the-shelf parts that could meet the slew requirements, so we tried the more conservative approach of fully passive I/V conversion followed by a passive 2-pole lowpass filter. At that point, any good audio-grade opamp or vacuum-tube gain stage could be used - and a 20 dB gain stage isn't rocket science to design.
Subjective results? Well, that was interesting. Conventional opamp-based I/V converters and active lowpass filters sounded like most CD players - fine most of the time, and "bad" CDs sounded - well, bad, the usual nasty congested early-CD sound. But passive conversion and lowpass filtering made the "bad" CD's perfectly listenable - they were bright and obviously mis-equalized in the recording studio - we suspected the old-school diameter-compensation for LP's had been left in the mastertape - but the congestion and distortion were all gone. In fact, most of the "digital" sound was gone as well, although the overall resolution of the CD was still obviously 16-bit, but at least we could hear all of it, and the annoying overhang of HF distortion was absent.
The 2000V/uSec video buffers? Well, they were all over the place subjectively, but we did find a few that sounded the same as the passive setup. So the active route was a possibility.
Matt and I are still amazed that almost no audio designers have performed our simple test, and almost none of them are aware of the ferocious requirements to avoid slewing during Nyquist reconstruction. Our best guess is that very few big-name audio designer have access to HP RF spectrum analyzers, and don't know, or don't care, about looking at audio equipment in the RF domain. Out of sight, out of mind. Most audio designers are trapped in a 20 Hz to 20 kHz world, and blithely assume if they can't see it on a 100 MHz scope, it isn't there.
Unfortunately, a scope is not the right tool for looking for low-level instabilities in power supplies, marginally unstable feedback systems, or seeing the full picture for digital devices. A scope will not trigger on an oscillation that is 20 dB down from the main signal (remember, we worked at Tektronix together, and we know what scopes can and cannot do). That's where a wideband spectrum analyzer is a necessity - you can peek down and look for tiny little items that are 60 dB down, and see them easily.
In fact, the Monarchy had an annoying 3 MHz oscillation that was 20 dB down - it was invisible on a scope, we looked - and it took all day to track it down to a 3-pin regulator that was used to down-regulate the LED display on the front panel. They had omitted the usual 100uF stabilization capacitor, and it was injecting noise into the ground system. That was a tricky little devil to find, but the solution was simple: disconnect the power supply to the regulator and the LEDs. The 3 MHz oscillation completely disappeared from sight. There's absolutely no way it could have been discovered, or resolved, with a scope. The audio industry is filled with products that don't work the way the designers intend, simply because XYZ famous-name audio company doesn't have the right tools for the job, or know how to use them. An RF spectrum analyzer is a must-have tool for digital work, or any amplifier with active regulation.
Sorry to digress from the subject of sibilance, but there are many design errors, or more generously, oversights in the audio chain that are responsible. Slewing is more common than generally realized in the audio-design community, and as mentioned earlier, many audiophile speakers have poorly designed crossovers. Laurie Fincham described the essentials of Target Function Design in the mid-Seventies, and I am surprised it is not universal practice some thirty years later.
Oddly enough, despite the ease of modern measurements and computer simulation (Laurie had to use a $150,000 DEC minicomputer, a full-time FORTRAN staff programmer, and a custom-built anechoic chamber), crossover design seems to have regressed in the last fifteen years, at least in the high-end. More speakers have 1st-order crossovers, crossover frequencies are moving downward, and rigid-cone midbass drivers with serious HF breakups are becoming more common. There seems to be some sort of industry-wide knowledge-loss going on.
Last edited:
Whoa.
Excellent information. Thank you.
I expected the square waves on the PCM63 to be fast... the "slewing holes" are interesting, but I'd expect that they would appear in the audio spectrum more as noise than anything else...
What would have been interesting to me would have been a comparison of in-band distortion between the passive I/V and the stock active, and the "improved" active with the highspeed video opamps...
The 3mhz. 3pin VR noise, ha ha.
The problem, I think is that wide bandwidth spectrum analyzers cost a fair amount of $$... I'd like to have one myself. Donations appreciated. But I agree completely, I'd want to be able to "see" out into the GHz. if I could...
But I do think this tale relates to "sibilance".
It certainly illustrates how "good" designs often have flaws that are on one hand "minor" but important in terms of the overall sound... often that difference coming in the higher freqs and the way they sound.
_-_-bear
Excellent information. Thank you.
I expected the square waves on the PCM63 to be fast... the "slewing holes" are interesting, but I'd expect that they would appear in the audio spectrum more as noise than anything else...
What would have been interesting to me would have been a comparison of in-band distortion between the passive I/V and the stock active, and the "improved" active with the highspeed video opamps...
The 3mhz. 3pin VR noise, ha ha.
The problem, I think is that wide bandwidth spectrum analyzers cost a fair amount of $$... I'd like to have one myself. Donations appreciated.
But I agree completely, I'd want to be able to "see" out into the GHz. if I could...But I do think this tale relates to "sibilance".
It certainly illustrates how "good" designs often have flaws that are on one hand "minor" but important in terms of the overall sound... often that difference coming in the higher freqs and the way they sound.
_-_-bear
Fourier analysis of square waves will indicate the faster the rise time, the more high frequency components will have higher amplitude. Theoretically, if the rise was infinitely fast, the harmonics would go out to infinity. A hf response peak would show up as overshoot on square waves and the amplitudes of HF components would be greater than Fourier predicts for square waves when seen on a spectrum analyzer. These components are not audible unless they cause intermodulation distortion to any component within the audible passband or result in overdriving some component beyond linearity. This first occurrance happend when a Japanese professor of electrical engineering tried to demonstrate to his students that harmonics above 20 khz were inaudible. He switched these harmonics in and out. But that was audible. The reason was discovered to be that they caused the tweeter to exhibit IM. When the speaker was redesigned with separate drivers for HF sound above and below 20 khz and a suitable crossover, the expected result was obtained and the ultrasonic components became inaudible. Unfortunately I don't have a reference for it.
Matt and I did a little asking around about the presence of slewing in the lowpass reconstruction - we were still in contact with our friends in the Tektronix digital groups, who were as knowledgeable as anyone I've ever met. Unfortunately, slewing at that critical portion of the DAC system results in a complex high-order distortion at the top of the audio band; it isn't noise at all, but distortion, and not benign 2nd-harmonic, either. As the stimulus frequency is decreased, the slewing still occurs, but there isn't as much, since the zero-crossings (where it occurs) aren't as frequent. In other words, 5 kHz had 1/4 as much as 20 kHz. Likewise, with decrease in stimulus magnitude, the slewing interval is decreased, although it is still happening around every zero crossing (one of the nastiest features of slewing).
In real terms, slewing in the reconstruction stage is akin to jitter in the digital path, which as we all know now, is only partially filtered by phase-lock-loop filtered. Back when I worked in the Tektronix Spectrum Analyzer Business Unit, we spent a lot of time chasing out "phase noise" from our analyzers, which is basically the same thing as jitter. Phase noise is not easy to filter; PLL's help, but when you're dealing with an analyzer with an on-screen dynamic range of 80 dB, a bandwidth going out to 1.8 GHz, and on-screen frequency resolution of 50 Hz, yes, phase noise is a big deal. We always tried to minimize it at the source, which was always much more effective than post-filtering.
DACs are defenseless against jitter, since that also violates Nyquist theory. The small differences in time between samples are translated into magnitude differences, since the overall width of the sample controls its energy level, as well as the assigned digital representation. In an ideal system, jitter collected at the time of initial conversion (which is always going to be greater than zero) should be made available during reconstruction to exactly cancel out the overall impact of jitter. But that would be so difficult I've never heard of anyone trying to do it.
So both jitter and slewing in the reconstruction phase have malign effects for PCM systems - both are similar to the effects of group-delay errors in FM transmission, where time errors are converted into amplitude distortion. The errors accumulate at higher frequencies (more transitions across the zero-crossing region) and at higher levels. This is exactly the opposite of many analog systems, where behavior around the zero-crossing is well-behaved (with the nasty exception of Class AB amplifiers).
Yes, it's true that Nyquist sampling is theoretically perfect. That's not the universe we live in. There are no Platonic ideals around here. The ideal of digital perfection is undermined by the inevitable presence of jitter - which again, is only partially filterable - and any slewing artifacts in reconstruction. Fortunately, the latter can be addressed with passive I/V conversion and passive 2-pole lowpass filtering. The good thing about passive conversion is that the distortion threshold for the I/V resistor (which has to be as perfect as possible) is pretty sharp; in the case of the PCM-63, there is no measurable improvement in going much below 100 ohms (10 and 100 ohms measured the same except for noise in the following stage). The correct value for this resistor is set by the "voltage compliance" of the ladder DAC, which is determined by experiment.
Delta-Sigma converters usually have built-in opamps, which conveniently masks the problem. As mentioned in the previous post, slewing distortion only slightly appears in spectral-analysis measurements, since the interval is so short compared to the main audio signal. But you do see it as a rise in HF distortion at the top of the band.
There are still a lot of people claiming "bits are bits". In what world is digital conversion (at both ends) done perfectly? There's the sweep-it-under-the-rug argument, which claims that XYZ distortion is "not audible" - well, because, followed by an appeal to authority. Not good enough. If there are errors in the system, first, they have to be thoroughly understood and analyzed, removed as much as possible, and then we can do subjective comparisons. To just arbitrarily say "it's not audible" without even bothering to examine the issue is not only lazy, but serves the marketers, who are always happy to sell low quality at a high price. (Lossy-compression downloads, anyone?)
In real terms, slewing in the reconstruction stage is akin to jitter in the digital path, which as we all know now, is only partially filtered by phase-lock-loop filtered. Back when I worked in the Tektronix Spectrum Analyzer Business Unit, we spent a lot of time chasing out "phase noise" from our analyzers, which is basically the same thing as jitter. Phase noise is not easy to filter; PLL's help, but when you're dealing with an analyzer with an on-screen dynamic range of 80 dB, a bandwidth going out to 1.8 GHz, and on-screen frequency resolution of 50 Hz, yes, phase noise is a big deal. We always tried to minimize it at the source, which was always much more effective than post-filtering.
DACs are defenseless against jitter, since that also violates Nyquist theory. The small differences in time between samples are translated into magnitude differences, since the overall width of the sample controls its energy level, as well as the assigned digital representation. In an ideal system, jitter collected at the time of initial conversion (which is always going to be greater than zero) should be made available during reconstruction to exactly cancel out the overall impact of jitter. But that would be so difficult I've never heard of anyone trying to do it.
So both jitter and slewing in the reconstruction phase have malign effects for PCM systems - both are similar to the effects of group-delay errors in FM transmission, where time errors are converted into amplitude distortion. The errors accumulate at higher frequencies (more transitions across the zero-crossing region) and at higher levels. This is exactly the opposite of many analog systems, where behavior around the zero-crossing is well-behaved (with the nasty exception of Class AB amplifiers).
Yes, it's true that Nyquist sampling is theoretically perfect. That's not the universe we live in. There are no Platonic ideals around here. The ideal of digital perfection is undermined by the inevitable presence of jitter - which again, is only partially filterable - and any slewing artifacts in reconstruction. Fortunately, the latter can be addressed with passive I/V conversion and passive 2-pole lowpass filtering. The good thing about passive conversion is that the distortion threshold for the I/V resistor (which has to be as perfect as possible) is pretty sharp; in the case of the PCM-63, there is no measurable improvement in going much below 100 ohms (10 and 100 ohms measured the same except for noise in the following stage). The correct value for this resistor is set by the "voltage compliance" of the ladder DAC, which is determined by experiment.
Delta-Sigma converters usually have built-in opamps, which conveniently masks the problem. As mentioned in the previous post, slewing distortion only slightly appears in spectral-analysis measurements, since the interval is so short compared to the main audio signal. But you do see it as a rise in HF distortion at the top of the band.
There are still a lot of people claiming "bits are bits". In what world is digital conversion (at both ends) done perfectly? There's the sweep-it-under-the-rug argument, which claims that XYZ distortion is "not audible" - well, because, followed by an appeal to authority. Not good enough. If there are errors in the system, first, they have to be thoroughly understood and analyzed, removed as much as possible, and then we can do subjective comparisons. To just arbitrarily say "it's not audible" without even bothering to examine the issue is not only lazy, but serves the marketers, who are always happy to sell low quality at a high price. (Lossy-compression downloads, anyone?)
Last edited:
It is critical to understand the limits of hearing. Designing beyond those limits for particular parameters has no legitimate purpose and can lead to degradation and other disadvantages for other parameters. There is actually no necessity for being able to reproduce sound above 20 khz because it is inaudible just as there is no purpose in designing photographic film or television sets to reproduce ultraviolet light.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.