Having 3-ways that are active and crossed with a DCX2496, my next move, once I have the sound perfected via measurements, is to implement the filters in analog. I'll then re-purpose the DCX for another use or possibly sell it.
To me, it is squandering the potential of an adjustable crossover on a static system.
I'll certainly buy this book when it comes out.
To me, it is squandering the potential of an adjustable crossover on a static system.
I'll certainly buy this book when it comes out.
I agree utterly with respect to very high slope filters. Even if one makes them subtractive, for example generating the highpass function by subtracting a linear-phase lowpass from an appropriately delayed fullrange feed, that will only sum properly at one location in the room, and most likely come a cropper everywhere else. Believe it or not, one can design FIRs with relatively shallow slopes which will behave well across a range of listening angles, which also has implications for a well managed reverberant field.
However, with respect I must disagree on linear-phase filters. Whereas I didn't hear a difference at 3 kHz crossover frequency, there was a change in the texture of percussion instruments at 250 Hz: snare drums are supposed to be rude, but minimum phase crossovers in that region makes them a tad overpolite, while going to linear phase restored the "eye-blink" quality of a proper snare drum (you blink when the drummer hits it).
I suspect that's why mini-monitors and fullrange speakers are seductive: even though the lower frequencies may be somewhat limited, there is better palpability of what lows do exist.
Quite likely. But is that the only audience that Douglas is trying to reach? Probably not.Quote:
Originally Posted by wintermute
Some might like the challenge of doing it the old way rather than taking the easy way out.
Absolutely not! I just got the feeling that the thread was taking on a tone of the analog stuff should be dropped and replaced with dsp stuff instead, going from one extreme to the other if you will
Was just trying to point out that whilst things might be moving forward, there is still an audience for the "old shcool" as well. Yes
I thought the first line in my post hinted at that Tony.
SoundEasy goes one better. It has a crossover emulation function which allows one to model a crossover (active or passive) and, via a multi-channel sound card plus amplifiers, listen to what it sounds like without having to build it... very tweak-friendly! There is at least one other software that has this capability (the name escapes me).
You are correct. Just bog standard slopes, tho a great choice of them. And a fair amount of EQ. But no bending of the slopes like it's easy to do with DIY analog filters.
So learning about the basic filter functions, Butterworth, Bessel, L-R, etc is a good start. Then learn how they can me modified to suit the speaker - in the analog or digital domain.
(That said, the short delay and phase adjustments of the DCX allow a lot of tweaking)
.
So learning about the basic filter functions, Butterworth, Bessel, L-R, etc is a good start. Then learn how they can me modified to suit the speaker - in the analog or digital domain.
(That said, the short delay and phase adjustments of the DCX allow a lot of tweaking)
.
The stumbling block is generally the delay used to correct for acoustic center misallignment and "steer the lobe" (a great feature of digital crossovers which is not commonly implemented with analog filters). The analog alternative is usually phase shift using asymetric slopes (at which point you've significantly deviated from the DCX "design").
The biggest weakness of the DCX is its limited ability to "correct" driver rolloff . . . which makes "bog standard" slopes (and phase shift) no longer "bog standard" at the acoustic output. More recent digital designs generally make that correction easy (with a bi-quad or two in the right places).
Now that will be DIY and interesting.
Should be. Current crossover setup is L-R 24db lowpass on the woofer, L-R 12db highpass and Butterworth 6db lowpass on the mid, and L-R 12db highpass on the tweeters. 3db, high Q notch at 223hz and 4db baffle step correction at 360hz. More fine tuning will happen when I get measurement gear setup.
By now it should be clear that a least the hobbyists are interested in DSP implementations of crossover filters. I would think the pros are, too.
He asked what was missing. There has been a resounding answer of "More DSP, please!" Seems pretty simple.
No?
The pros have been playing with DSP for a while now: Sony showed off a DSP corrected and crossed-over speaker at least a decade ago, and John Dunlavy wanted to bring out a version.
Problem is, who would buy such a thing? The extra power amps increase cost for relatively little benefit in a world where MP3 is the be-all of sound quality. If you do target serious [1] audiophiles, many of them are vinyl aficionados who consider anathema all things digital and who would view digitisation with deep suspicion. DSP speakers occupy a narrow niche, probably best marketed as an extremely high end lifestyle product to people with combined AV and music rigs. Even NHT gave up after people couldn't see spending $6000 on an excellent subwoofer/mini-monitor system, and that bad boy was designed by Jack Hidley. Meridian, bless their hearts, seem to be the only ones who know how to sell digital to the punters.
[1] as opposed to Mpingo disc polishing woo-woo addicts
So it looks as if hobbyists are the main ones carrying forward speaker DSP, at least until someone comes up with a $1.98 power amp.
As to whether DS should add DSP to his book, that'd be a GREAT idea if he has another 480 pages for the subject. What kind of filters does one use? IIR? Then you must consider filter design methods such as bilinear transform, what happens to poles & zeros formerly in the left hand plane, word-length effects for fixed-point words, or in floating point whether single or double precision is needed. FIR? First of all, FIRs have no correspondance to the analog world so they have to be explained. Then what design method is used? Parks-McClellan, maximally flat, windowed sinc? Each one has its advantages in different audio contexts.
Self talks about various opamps in his book, so to cover digital similarly he would have to discuss the merits of different ADC/DACs, fixed and floating point math (hint: single precision isn't necessarily better than 32 bit fixed point, because the mantissa is only 24 bits), different filter structures for IIR, and convolution methods for FIR, including FFTs.
DSP, like analog, includes many subtleties if one wants to do it well. Slapping a couple of chapters on the end of a very good analog book would not do the subject justice.
What about distributed LC transmission lines? Pretty similar, I would say. In fact, it would be really cool if someone would make such a filter, just for the fun of it. Can achieve really sharp filtering at HF.
--
Greetz,
MatchASM
I assume you already have the power amps in place and functional, just waiting for a crossover to feed them. Considering your heavy schedule, what sort of time frame are you thinking of to get this project to a first audition, maybe even a thread?
Hi,
Which is why my latest speaker has a first order series crossover.
Took several weeks "to work" train journeys in PCD to get a first cut (make driver models and all). Then messing a lot with values I found that things tracked quite well giving a huge degree of tolerance before getting bad... The practical implementation too around 2 Hours...
With a digital crossover and more amp's it would have taken a lot less thyme (and parsley and sage and rosemarine), but I do not have three channels of good (not blameless) amplifiers at hand right now...
So always horses for courses.
I may have overemphasised digital solutions above. The point was not that no-one uses analog anymore, just that it is SO OLD (timecoherent fully passive 2-Way systems with good squarewave performance where available in the late 1930's) that writing reference works only about them seems like a sure shelffiller. At the bookshop.
Ciao T
Considering your heavy schedule, what sort of time frame are you thinking of to get this project to a first audition, maybe even a thread?
Which is why my latest speaker has a first order series crossover.
Took several weeks "to work" train journeys in PCD to get a first cut (make driver models and all). Then messing a lot with values I found that things tracked quite well giving a huge degree of tolerance before getting bad... The practical implementation too around 2 Hours...
With a digital crossover and more amp's it would have taken a lot less thyme (and parsley and sage and rosemarine), but I do not have three channels of good (not blameless) amplifiers at hand right now...
So always horses for courses.
I may have overemphasised digital solutions above. The point was not that no-one uses analog anymore, just that it is SO OLD (timecoherent fully passive 2-Way systems with good squarewave performance where available in the late 1930's) that writing reference works only about them seems like a sure shelffiller. At the bookshop.
Ciao T
Hi,
Concerning audio, a great part of what I know and what I built comes from my reading of british Electronics World (EW). The first issue I bought was the December 1978. The TOC contents P. Baxandall and S. Linkwitz as authors, there could'nt be better baptismal fonts of clever audio for me.
Apart from some amplifiers (there is a blameless, of course) and two major Cyril Bateman's projects, I built :
- a Hardman Crossover (EW August 1999).
DiyAudio related thread :
http://www.diyaudio.com/forums/multi-way/6655-best-active-crossover-here.html
It was quite difficult to find the initial Bainter's paper which inspired it, the link mentionned in the above thread is not good anymore. If interested, contact me.
- a Sokol-Hegglun transform circuit (Sokol, EW December 1983; Hegglun, EW May 1996), which does the same job as a Linkwitz transform but has the desirable feature of being adjustable.
DiyAudio related post :
http://www.diyaudio.com/forums/mult...r-their-resonant-frequency-2.html#post1786742
Both gave me excellent results. Wanting to better master everything, I began to use Spice simulators. Some time later, I became convinced that digital technology can give excellent sound, so I looked for a digital crossover processor. I purchased a model which is considered as one of the best of the pro market and which includes an NTM crossover which mimics the elliptic Hardman's circuit.
The following shows that a good knowledge of analog circuits is of great help even if you do not use them for crossovers. I also present some unconventionnal designs aiming for good phase/group delay responses.
Crossovers and driver transfer functions
An important aspect of crossovers is that they should include the tranfer functions of the loudspeakers. This was done in EW December 1978 Linkwitz's project, using a transform to increase the resonance of the medium and tweeter transfer functions (the other way of transform usage, better known, is to decrease the resonance of the woofer transfer function). Other examples are indeed very rare.
Area transforms can be done using a standard digital processor having parametric and equalisation corrections ? This is where simulation using analog circuits come into play and helps a lot. I do not think I've seen it mentionned elsewhere but I soon found that parametric corrections can rigorously change the apparent Q of the resonance of a driver having a 2nd order transfer function.
To increase the Q of a driver, the parametric corrections must be set, at the resonant frequency, to the target Q and the gain, positive, to the dB difference between the target Q and the initial Q.
To decrease the Q of a driver, the parametric corrections must be set at the resonant frequency to the initial Q and the gain, negative, to the dB difference between the target Q and the initial Q.
I have not made any calculations, this is what the simulator showed me. Here Q is the Qtc of the driver.
Beware that among different digital processors, values of Q are not made equal to get an identical effect. BSS and Yamaha use the same as the Rane definition, XTA and Ram Audio do not :
Bandwidth in Octaves Versus Q in Bandpass Filters
Why DSP Boxes Set the Same Way Differ
Mastering the Q as desired, shelves are now needed to get a new apparent resonant frequency. This is where the greatest difficulties rely because algorithms of "shelves" slopes of digital processors do not sufficiently follow the curves obtained by simple analog circuits using R, C and op-amps components.
I had to import the processor's curves measured with an Arta-like software and a sound card hardware to the simulator. My processor has a 12 dB slope shelf option with which I can get a replica of an analog Linkwitz transform if the initial and target Qs are adequately set.
However there is another solution, quite elegant, once more inspired by some EW authors : make a 2nd order filter looking as having a 6 dB/o slope. To do so set the Q of the resonance very low, under 0.5. Then use a 6 db/o shelf equalisation to get the aimed frequency response. Once again, simulation helps to determine it taking in account the 6 dB/o shelf curve imported from the processor. I could achieve curves with less than 0.3 dB deviations from the theorical results.
Another way to achieve low Q is to use some form of loudspeaker velocity feedback which lowers distorsion.
Quasi-optimal filters
There is a family of crossovers, named quasi-optimal, which is quite popular in France. The idea is to get the best compromise between axis frequency response, group delay and what is called "peaking axis response" by Rane, "coincidence response" in France and which is calculated by summing the modules of the frequency response of each way.
The most famous is the Jean-Michel Le Cleac'h's crossover using separated frequencies and delay in the high pass way. For fx = crossover frequency :
Le Cleach's quasi-optimal crossover
high pass = Butterworth 18 dB/o at 1.1456 * fx, phase inverted, time delay = 0.22 / fx
low pass = Butterworth 18 dB/o at 0.8729 * fx
Documentation in french, many figures :
filtrage
Documentation in english :
http://freerider.dyndns.org/anlage/LeCleach2.zip
Expanded Soundstaging and 3D-Imaging
Another quasi-optimal is proposed by Francis Brooke :
Brooke's quasi-optimal crossover
high pass : Butterworth 12 dB/o at fx + Butterworth 6 dB/o at at 0.976 * fx, phase inverted, time delay = 0.21/fx
low pass : Butterworth 12 dB/o at fx + Butterworth 6 dB/o at 1.025 * fx
Documentation in french, many figures :
http://francis.audio2.pagesperso-orange.fr/Quasi_Linkwitz_ordre3.pdf
It was only after a third try that I became convinced by this kind of filters, I think because my system does now suffer much less of diffracion than it used to be. See picture below how I minimised diffraction.
I now have abandoned the NTM in favour of a Brooke with compensation of the high pass transfer function of the drivers.
Low pass function of drivers
I have made some measurements of drivers to determine if their low-pass function could be taken in account in crossovers, which is quite difficult because the slope is often 24 dB/o or even more. I submitted the results to Jean-Michel and Francis. Our conclusions are that loudspeakers at their end of band-pass zone can relatively well be considered as minimal phase systems.
It has been suggested to me not to try any frequency response correction of the low-pass function but to add a little delay. I've not simulated this yet.
Gerd Schmidt
To end this post, I would like mention two EW's "circuit ideas", both by Gerd Schmidt, with the aim to get linear phase response :
July 1999, page 572 (see also "Letters", EW November 1997, page #957) :
"New FIR phase crossover"
April 2001 issue, page #294,
"A switchable constant power or linear loudspeaker crossover"
based on Yamnaka and Baekgaard (Bang and Olufsen) works.
I've not simulated them yet.
Beware that searching links to "Gerd Schmidt" on the net drives to a false antivirus AV8.
.
Concerning audio, a great part of what I know and what I built comes from my reading of british Electronics World (EW). The first issue I bought was the December 1978. The TOC contents P. Baxandall and S. Linkwitz as authors, there could'nt be better baptismal fonts of clever audio for me.
Apart from some amplifiers (there is a blameless, of course) and two major Cyril Bateman's projects, I built :
- a Hardman Crossover (EW August 1999).
DiyAudio related thread :
http://www.diyaudio.com/forums/multi-way/6655-best-active-crossover-here.html
It was quite difficult to find the initial Bainter's paper which inspired it, the link mentionned in the above thread is not good anymore. If interested, contact me.
- a Sokol-Hegglun transform circuit (Sokol, EW December 1983; Hegglun, EW May 1996), which does the same job as a Linkwitz transform but has the desirable feature of being adjustable.
DiyAudio related post :
http://www.diyaudio.com/forums/mult...r-their-resonant-frequency-2.html#post1786742
Both gave me excellent results. Wanting to better master everything, I began to use Spice simulators. Some time later, I became convinced that digital technology can give excellent sound, so I looked for a digital crossover processor. I purchased a model which is considered as one of the best of the pro market and which includes an NTM crossover which mimics the elliptic Hardman's circuit.
The following shows that a good knowledge of analog circuits is of great help even if you do not use them for crossovers. I also present some unconventionnal designs aiming for good phase/group delay responses.
Crossovers and driver transfer functions
An important aspect of crossovers is that they should include the tranfer functions of the loudspeakers. This was done in EW December 1978 Linkwitz's project, using a transform to increase the resonance of the medium and tweeter transfer functions (the other way of transform usage, better known, is to decrease the resonance of the woofer transfer function). Other examples are indeed very rare.
Area transforms can be done using a standard digital processor having parametric and equalisation corrections ? This is where simulation using analog circuits come into play and helps a lot. I do not think I've seen it mentionned elsewhere but I soon found that parametric corrections can rigorously change the apparent Q of the resonance of a driver having a 2nd order transfer function.
To increase the Q of a driver, the parametric corrections must be set, at the resonant frequency, to the target Q and the gain, positive, to the dB difference between the target Q and the initial Q.
To decrease the Q of a driver, the parametric corrections must be set at the resonant frequency to the initial Q and the gain, negative, to the dB difference between the target Q and the initial Q.
I have not made any calculations, this is what the simulator showed me. Here Q is the Qtc of the driver.
Beware that among different digital processors, values of Q are not made equal to get an identical effect. BSS and Yamaha use the same as the Rane definition, XTA and Ram Audio do not :
Bandwidth in Octaves Versus Q in Bandpass Filters
Why DSP Boxes Set the Same Way Differ
Mastering the Q as desired, shelves are now needed to get a new apparent resonant frequency. This is where the greatest difficulties rely because algorithms of "shelves" slopes of digital processors do not sufficiently follow the curves obtained by simple analog circuits using R, C and op-amps components.
I had to import the processor's curves measured with an Arta-like software and a sound card hardware to the simulator. My processor has a 12 dB slope shelf option with which I can get a replica of an analog Linkwitz transform if the initial and target Qs are adequately set.
However there is another solution, quite elegant, once more inspired by some EW authors : make a 2nd order filter looking as having a 6 dB/o slope. To do so set the Q of the resonance very low, under 0.5. Then use a 6 db/o shelf equalisation to get the aimed frequency response. Once again, simulation helps to determine it taking in account the 6 dB/o shelf curve imported from the processor. I could achieve curves with less than 0.3 dB deviations from the theorical results.
Another way to achieve low Q is to use some form of loudspeaker velocity feedback which lowers distorsion.
Quasi-optimal filters
There is a family of crossovers, named quasi-optimal, which is quite popular in France. The idea is to get the best compromise between axis frequency response, group delay and what is called "peaking axis response" by Rane, "coincidence response" in France and which is calculated by summing the modules of the frequency response of each way.
The most famous is the Jean-Michel Le Cleac'h's crossover using separated frequencies and delay in the high pass way. For fx = crossover frequency :
Le Cleach's quasi-optimal crossover
high pass = Butterworth 18 dB/o at 1.1456 * fx, phase inverted, time delay = 0.22 / fx
low pass = Butterworth 18 dB/o at 0.8729 * fx
Documentation in french, many figures :
filtrage
Documentation in english :
http://freerider.dyndns.org/anlage/LeCleach2.zip
Expanded Soundstaging and 3D-Imaging
Another quasi-optimal is proposed by Francis Brooke :
Brooke's quasi-optimal crossover
high pass : Butterworth 12 dB/o at fx + Butterworth 6 dB/o at at 0.976 * fx, phase inverted, time delay = 0.21/fx
low pass : Butterworth 12 dB/o at fx + Butterworth 6 dB/o at 1.025 * fx
Documentation in french, many figures :
http://francis.audio2.pagesperso-orange.fr/Quasi_Linkwitz_ordre3.pdf
It was only after a third try that I became convinced by this kind of filters, I think because my system does now suffer much less of diffracion than it used to be. See picture below how I minimised diffraction.
I now have abandoned the NTM in favour of a Brooke with compensation of the high pass transfer function of the drivers.
Low pass function of drivers
I have made some measurements of drivers to determine if their low-pass function could be taken in account in crossovers, which is quite difficult because the slope is often 24 dB/o or even more. I submitted the results to Jean-Michel and Francis. Our conclusions are that loudspeakers at their end of band-pass zone can relatively well be considered as minimal phase systems.
It has been suggested to me not to try any frequency response correction of the low-pass function but to add a little delay. I've not simulated this yet.
Gerd Schmidt
To end this post, I would like mention two EW's "circuit ideas", both by Gerd Schmidt, with the aim to get linear phase response :
July 1999, page 572 (see also "Letters", EW November 1997, page #957) :
"New FIR phase crossover"
April 2001 issue, page #294,
"A switchable constant power or linear loudspeaker crossover"
based on Yamnaka and Baekgaard (Bang and Olufsen) works.
I've not simulated them yet.
Beware that searching links to "Gerd Schmidt" on the net drives to a false antivirus AV8.
.
Attachments
@ forr: very interesting! But where have you been the past 15+ years? A lot of what you speak of can be designed graphically using the tools freely available from the FRD Consortium, Speaker Workshop or this free design from GroundSound (it only works with their own DSP boards but is interesting to play with): XOverWizard
NB: I'm not disrespecting the learning you did, but the methods employed are easier accessible with modern tools.
NB: I'm not disrespecting the learning you did, but the methods employed are easier accessible with modern tools.
Last edited:
I assume you already have the power amps in place and functional, just waiting for a crossover to feed them. Considering your heavy schedule, what sort of time frame are you thinking of to get this project to a first audition, maybe even a thread?
The 6 channel amp has been in operation for nearly 2 years, with a DCX2496 as the crossover. Currently, I'm working on a new power supply for the unit ( see HERE ). Once that is complete, I'll start the crossover project.
Could be a while yet...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.