I edited the html and changed the background color; can now see the graphs.
There is no reason not to model the drivers as point sources if they are implemented in a reasonable manner, i.e., radiated wavelenght is large compared to piston diameter. (KA=1 breakpoint).
The off-axis responses in these graphs are one of the reasons why odd-order BUT are desirable. If the listener moves off-axis this way, response appraoches only +3 dB peak; if moves that way, approaches infinite dip which is far more audible than a small peak. If even-order BUT is used, moving either this way or that causes the response to approach an infinite dip. With odd-order the single infinite dip is somewhat ameliorated by constant total power into the room, but to a much lesser extent with even-order due to having 2 dips. With Linkwitz-Riley (even-order), the total power into the room already has a dip, so the effect is even worse. There is no way to diminish this effect without sacrificing flat total power. While constant total power is of lesser importance than flat axis power, it still cannot be ignored; the effect of total power varies with frequency, becoming increasingly audible at lower frequencies.
All Butterworth filters tuned to the same cutoff frequency exhibit constant total power when HP anf LP are summed.
Hence, odd-order BUT is the logical choice, but should not be greater than 3rd. Psychoacoustic studies have shown that excessive phase shift across the spectrum from higher-order filters is most definitely audible and undesirable. Properly implemented 3rd order BUT where adjacent drivers are electrically of opposite polarity (still has constant power and voltage) exhibit only 180 deg. total phase shift, being only 90 deg at xover point. With LR4 (adjacent drivers have same polarity), total shift is 360 deg, being 180 deg at xover point. 360 deg IS NOT the same as 0 deg - watch on a scope how the phase of summed filters starts at 0, proceeds to 180 then continues on to 360 (NOT back to 0).
There is no reason not to model the drivers as point sources if they are implemented in a reasonable manner, i.e., radiated wavelenght is large compared to piston diameter. (KA=1 breakpoint).
The off-axis responses in these graphs are one of the reasons why odd-order BUT are desirable. If the listener moves off-axis this way, response appraoches only +3 dB peak; if moves that way, approaches infinite dip which is far more audible than a small peak. If even-order BUT is used, moving either this way or that causes the response to approach an infinite dip. With odd-order the single infinite dip is somewhat ameliorated by constant total power into the room, but to a much lesser extent with even-order due to having 2 dips. With Linkwitz-Riley (even-order), the total power into the room already has a dip, so the effect is even worse. There is no way to diminish this effect without sacrificing flat total power. While constant total power is of lesser importance than flat axis power, it still cannot be ignored; the effect of total power varies with frequency, becoming increasingly audible at lower frequencies.
All Butterworth filters tuned to the same cutoff frequency exhibit constant total power when HP anf LP are summed.
Hence, odd-order BUT is the logical choice, but should not be greater than 3rd. Psychoacoustic studies have shown that excessive phase shift across the spectrum from higher-order filters is most definitely audible and undesirable. Properly implemented 3rd order BUT where adjacent drivers are electrically of opposite polarity (still has constant power and voltage) exhibit only 180 deg. total phase shift, being only 90 deg at xover point. With LR4 (adjacent drivers have same polarity), total shift is 360 deg, being 180 deg at xover point. 360 deg IS NOT the same as 0 deg - watch on a scope how the phase of summed filters starts at 0, proceeds to 180 then continues on to 360 (NOT back to 0).
Last edited:
Universal filter board update...
OK, getting back to the topic of active crossover PCBs and kits...
Over the weekend I completed the preliminary layout work on the "universal filter board" PCB. The schematic of the board is found in the pdf file at the end of this post.
There are two sections per board, and each section can be wired as on of the following:
inverting and non-inverting gain stage
1st order HP or LP filter stage
2nd order HP or LP filter stage (MFB or Sallen-Key)
2nd order filter stage plus notch, HP or LP
notch filter
3rd order HP or LP filter stage (MFB or Sallen-Key)
first or second order all-pass filter
Each of the sections uses a single op amp, so amplifier related noise and distortion issues will be kept to a minimum. Both sections allow for an RC filter to be located at the output of the op-amp. It is recommended that an RC low-pass filter be placed here to fix issues that are inherent in Sallen-Key or MFB LP stages, in which the amount of attenuation decreases at high frequencies (which allows noise to be passed by the filter block). This also will provide some isolation of the op-amp from RF frequency that a long cable run can pick up.
DESIGN NOTES:
In order to make the board as general as possible, so that all of these configurations can be created, each position on the board must be able to accommodate a cap or a resistor. As a result, each position has to be sized to allow for the largest component to fit there. This seems to be the caps. Checking through some datasheets for 1% and 2% tolerance PP caps from WIMA and Vishay, it seems that a footprint of about 10mm x 18mm (WxL) will allow caps up to about 1uF to fit on the board. There shouldn't strictly be a need to use a cap this large, but it does give some extra flexibility in scaling the impedance of all the components in the circuit.
To allow for different component pin spacings, each position on the board will have thru-holes that accommodate 5mm, 10,, and 15mm pin spacing, plus possibly 7.5mm (which seems to be standard for some lines of caps).
In order to fit 2 circuits worth of all of these large components it looks like I will have to increase the board footprint to about 4"x4", however I plan to provide mounting hole locations that will line up with the other boards I am designing, which are all on a 2"x4" footprint.
So far this is coming together nicely. Hopefully I can finish off the PCB design in the next day or two. The Fab re-opens in a couple of days, so I should be sending off orders within the next week for sure.
-Charlie
The schematic for one section of the universal filter board is shown in the linked pdf doc, below:
UNIVERSAL FILTER BOARD CIRCUIT CONNECTIONS
OK, getting back to the topic of active crossover PCBs and kits...
Over the weekend I completed the preliminary layout work on the "universal filter board" PCB. The schematic of the board is found in the pdf file at the end of this post.
There are two sections per board, and each section can be wired as on of the following:
inverting and non-inverting gain stage
1st order HP or LP filter stage
2nd order HP or LP filter stage (MFB or Sallen-Key)
2nd order filter stage plus notch, HP or LP
notch filter
3rd order HP or LP filter stage (MFB or Sallen-Key)
first or second order all-pass filter
Each of the sections uses a single op amp, so amplifier related noise and distortion issues will be kept to a minimum. Both sections allow for an RC filter to be located at the output of the op-amp. It is recommended that an RC low-pass filter be placed here to fix issues that are inherent in Sallen-Key or MFB LP stages, in which the amount of attenuation decreases at high frequencies (which allows noise to be passed by the filter block). This also will provide some isolation of the op-amp from RF frequency that a long cable run can pick up.
DESIGN NOTES:
In order to make the board as general as possible, so that all of these configurations can be created, each position on the board must be able to accommodate a cap or a resistor. As a result, each position has to be sized to allow for the largest component to fit there. This seems to be the caps. Checking through some datasheets for 1% and 2% tolerance PP caps from WIMA and Vishay, it seems that a footprint of about 10mm x 18mm (WxL) will allow caps up to about 1uF to fit on the board. There shouldn't strictly be a need to use a cap this large, but it does give some extra flexibility in scaling the impedance of all the components in the circuit.
To allow for different component pin spacings, each position on the board will have thru-holes that accommodate 5mm, 10,, and 15mm pin spacing, plus possibly 7.5mm (which seems to be standard for some lines of caps).
In order to fit 2 circuits worth of all of these large components it looks like I will have to increase the board footprint to about 4"x4", however I plan to provide mounting hole locations that will line up with the other boards I am designing, which are all on a 2"x4" footprint.
So far this is coming together nicely. Hopefully I can finish off the PCB design in the next day or two. The Fab re-opens in a couple of days, so I should be sending off orders within the next week for sure.
-Charlie
The schematic for one section of the universal filter board is shown in the linked pdf doc, below:
UNIVERSAL FILTER BOARD CIRCUIT CONNECTIONS
3rd Order Butterworth w/ Time-Alignment
Time-Alignment is a goal to be achieved all unto itself. It is possible to stagger drivers such that one is displaced by one wavelength at crossover, which yields flat measured response. (But this creates small comb-filter anaomolies above and below xover). Another psychoacoustic effect comes into play as the drivers are moved into correct time-alignment that is not related to frequency response. The human ear can, in some people, detect the improved realism that occurrs when the energy from all the drivers is coherent. This design has been mostly overlooked because of it's extra cost to implement and maybe unpleasing cosmetic appeaarance. Any serious audiophile owes it to himself to at least do the experiment.
Looks like your filter circuit topolgy has covered all the bases. Do you plan to offer additional stage with LFEQ to compensate for 2nd-order sealed-box or infinite baffle rolloff? Linkwitz-Riley has presented such a circuit, and although it differs in topology from Auratron, it will yield correct results. Check their website for spreadsheet calculator to download.
Time-Alignment is a goal to be achieved all unto itself. It is possible to stagger drivers such that one is displaced by one wavelength at crossover, which yields flat measured response. (But this creates small comb-filter anaomolies above and below xover). Another psychoacoustic effect comes into play as the drivers are moved into correct time-alignment that is not related to frequency response. The human ear can, in some people, detect the improved realism that occurrs when the energy from all the drivers is coherent. This design has been mostly overlooked because of it's extra cost to implement and maybe unpleasing cosmetic appeaarance. Any serious audiophile owes it to himself to at least do the experiment.
Looks like your filter circuit topolgy has covered all the bases. Do you plan to offer additional stage with LFEQ to compensate for 2nd-order sealed-box or infinite baffle rolloff? Linkwitz-Riley has presented such a circuit, and although it differs in topology from Auratron, it will yield correct results. Check their website for spreadsheet calculator to download.
2nd-order EQ Follow-up
Here is link to Linkwitz-Riley EQ :
http://sound.westhost.com/project71.htm
Auratron implements this with a single stage, but they do not offer details.
This EQ works great for sealed-box mid-bass; eq needs to extend at least 1, better 2, octaves below xover.
Here is link to Linkwitz-Riley EQ :
http://sound.westhost.com/project71.htm
Auratron implements this with a single stage, but they do not offer details.
This EQ works great for sealed-box mid-bass; eq needs to extend at least 1, better 2, octaves below xover.
Last edited:
Linkwitz (not Riley) invented the circuit and published it in Speaker Builder vol. 2, 1980. That's the standard that other implementations (such as the Auratron company you mention) are judged, in terms of "correct results", not the other way around.
I've developed a circuit that does the same thing as the LT (Sigfried's single op amp circuit) but does not have any limitations (like his "k must be greater than zero" restriction) on what transfer functions can be implemented. Like the state-variable filter board I am developing, after you build it you can make a wide range of adjustments, so you could adjust the bass response for different listening spaces for instance, or re-use the board in another project. I plan to offer it as a kit of PCB plus parts.
It's important to keep in mind that the LT is much more than a low frequency equalizer, or only for subwoofers. You can use it on any driver in a sealed enclosure, even on a tweeter, to change the response in to a new one. SL talks about the latter application in his web page(s) about the LT. If there is any location for a good reference on the LT, it's there.
Say, why do you keep making references to that horrible web site of Auratron. I can't stand to look at the pages, with all the neon colored text-it hurts my eyes! It's like VGA graphics from 1975 gone bad.
-Charlie
The ESP you linked to also uses a single stage for the LT circuit. Check again. It's U2 in Figure 2. The other op amps (U1A and U1B) provide mixing of L and R channels, a HP filter to remove DC, and a phase reversal option. Does the Auratron offer these options in their circuit?
Again, note that Riley had nothing to do with the LT. For your reference, here is a link to his 1980 paper:
http://www.linkwitzlab.com/x-sb80-3wy.htm
-Charlie
Last edited:
I was referring to LW transform to be used as an inverse 2nd order filter like the one you have already developed, not for subwoofer eq.
Auratron color scheme is weird. I refer to them because all the info is presented there in succinct form, and I know from personal experience that their approach works. There probably are other companies that offer similar design approaches, that just happens to be the one I know.
Auratron color scheme is weird. I refer to them because all the info is presented there in succinct form, and I know from personal experience that their approach works. There probably are other companies that offer similar design approaches, that just happens to be the one I know.
Thanks for the clarification and new links. I don't believe Auratron xover was meant to be versatile. The one I have only has provision for 3rd order BUT with optional LFEQ.
All of my converstions with you on this thread were to help insure all reasonable options are covered by your "universal xover" and that interested parties are aware of all the various implementations and options. Thanks for considering all my input.
All of my converstions with you on this thread were to help insure all reasonable options are covered by your "universal xover" and that interested parties are aware of all the various implementations and options. Thanks for considering all my input.
Crossover Design
Check this reference :
Small, R. H., "Constant-Voltage Crossover Network Design" JAES Vol 19, no 1, Jan-1971
Check this reference :
Small, R. H., "Constant-Voltage Crossover Network Design" JAES Vol 19, no 1, Jan-1971
I'm not currently subscribed to the AES, so I would have to make a trip to the library. I believe that this is Small's original paper on the Butterworth crossover. Is that correct?
You might want to take a look at this Rane app note (available online):
THis document provides a good overview of Butterworth and Linkwitz-Riley crossovers, and lists some advantages and disadvantages of each. Note that Figure 1b is exactly like the model of off-axis response of the Butterworth 3rd order Xover that I posted. Figure 1a shows the polar pattern (just a different way to plot the same data) at the crossover frequency. As you can see, above the on axis plane there is a peak of 3dB and below a null. With the 3rd order Butterworth, if you reverse the driver phase, the position of the null and peak switches, but the pattern is the same "shape". This is the "problematic" behavior that I was talking about for the BUT3 crossover in a previous post. These figures just show it a little more clearly.
Because the off axis nulls for the LR4 are symmetric about the on-axis position, one can easily boost this band of frequencies. In my modeling, the "null" for LR4 is only partial, unlike what is shown in Figure 2 of the Rane app note, about -3dB. This is easily compensated by an EQ boost of only 2dB, so that on and off-axis is within 1-2dB of flat. This will restore the power response to near flat, while keeping on and off axis frequency response also nearly flat. This seems like a good compromise between those two response metrics.
It is important to keep in mind that the graphs and analysis presented in the Rane app note assume that the pathlength between the drivers and the listening position is exactly the same, e.g. that they have been "time aligned". Unless that is true, the patterns can shift substantially in very odd ways that can result in on axis nulls or other problems. Time alignment, or a very careful consideration of the summed driver response at the design stage and careful measurements after the build, are necessary to ensure that the response is smooth, both on and off axis, for all frequencies in the crossover region. It'e pretty straightforward to do time alignment using all-pass filters, but modeling the system and how the all-pass stages are used is key to pulling off a great design.
-Charlie
Thanks for joining in the discussion and giving your input. I think that your suggestion to allow for a 3rd order single-amplifier section was a great one! I am glad you brought it up, just in time for the design of the board.
-Charlie
Check back to posts # 81 and # 83
I do not consider the asymmetry of BUT3 to be problematic, but rather an asset. The symmetry of the infinite dips means no matter which way listener moves, response will more rapidly approach a large dip.
If EQ is added so off-axis dips are improved, on-axis response will no longer be flat. (The very nature of stereo implies that there will always be a "sweet spot".)
Electrical phase-shift netrworks do not yield the same effect as mechanical alignment of the drivers. Looks good on paper, but does not pass the listening test.
Based on the analysis, in my opinion, BUT3 w/ mechanically time-aligned drivers represents the ultimate sonic experience. Maybe other listeners will agree?
I do not consider the asymmetry of BUT3 to be problematic, but rather an asset. The symmetry of the infinite dips means no matter which way listener moves, response will more rapidly approach a large dip.
If EQ is added so off-axis dips are improved, on-axis response will no longer be flat. (The very nature of stereo implies that there will always be a "sweet spot".)
Electrical phase-shift netrworks do not yield the same effect as mechanical alignment of the drivers. Looks good on paper, but does not pass the listening test.
Based on the analysis, in my opinion, BUT3 w/ mechanically time-aligned drivers represents the ultimate sonic experience. Maybe other listeners will agree?
Well, it looks like we will just have to disagree again.
If you want to continue to hash this out, I suggest that we start another thread on the topic, with a title like "Butterworth 3rd order versus LR4 - which is best?" or something catchy. Maybe that would attract more input from others?
I'd like to try and get this thread back on topic, e.g. about crossover board design, and options, and about other circuits related to loudspeaker crossovers. I don't have a problem discussing the merits BUT3 and LR4, etc. but I was hoping this thread would be more about how to implement those circuits, not about which one is superior. That is up to each designer to decide - IMHO it's a personal decision.
What do you think about that approach?
-Charlie
I agree. However, I believe we have covered all the technical aspects, so now it is up to the listener to decide.
About circuit implementation, I also suggested that, where possible, single-stage inverting was optimum, and looking back at previous posts, I think you agreed. Based on the circuit description a few posts back, I don't see any room for improvement. If you plan to add gain control, I suggest switched resistor networks.
About circuit implementation, I also suggested that, where possible, single-stage inverting was optimum, and looking back at previous posts, I think you agreed. Based on the circuit description a few posts back, I don't see any room for improvement. If you plan to add gain control, I suggest switched resistor networks.
Are you referring to a stepped-attenuator type circuit, using relays? Not sure...
Again, thanks for all of your suggestions on the circuit(s).
-Charlie
No relays - toooo big. Check this link:
Control Switches | Diy HiFi Supply
Use 1% resistors. Here is another link :
http://users.ece.gatech.edu/~mleach/papers/stepatten/stepatten.pdf
Marshall Leach passed away last month.
These don't need a circuit board, you just wire everything directly to the stepped attenuator. Series and shunt wiring options are available and you can buy them premade in various places, including Ebay. YMMV.
I have seen some good examples of the same thing, except that instead of mechanically switching between contacts, miniature relays are used. Depending on how you wire the circuit, you can minimize the number of relays, and the miniature types have reliable contacts and switch very fast. If you implement the resistor attenuation network as a "tree" you can do a 2^N type of arrangement that give you, say 64 or 128 levels of attenuation with 7 or 8 relays. You can't touch that with one of these mechanical stepped attenuators. Here is one example that is out there:
http://www.vaneijndhoven.net/jos/switchr/design.html
I saw a nice implementation of this using an Arduino front end at the last Burning Amp festival in San Francisco. That kind of thing I could probably do as a PCB, but I would need to dream up a different front end implementation, since I tend to avoid working with digital circuits. One has to be careful about switching transients, but this can be addressed. You can also use a DAC chip to implement a stepped attenuator, since the internal resistor network can do a similar thing.
In the past I considered building an active 6-channel gain control based on a VGA like the SSM2018T or equivalent ICs (e.g. from Cool Audio), but these don't have top notch distortion, being on par with some of the now popular LDR based attenuators, so I never went through with it.
-Charlie
Last edited:
There are no active switches, i.e. digital, that don't have unacceptably high distortion. Avoid them at all cost. The switched resistor or relay implementation will allways offer superior performance. The problem with them is that as the analog signal goes thru 0. the resistance varies, even for a constant control voltage. This phenomenon is the very definition of distortion.
Multiple boards
Can 2 or more boards be used in series to create higher-order / steeper slope configurations?"
Can 2 or more boards be used in series to create higher-order / steeper slope configurations?"
Yes, that's the way to do it. Each section can be configured as first or second order filter and then cascaded (connected in series) to create a higher order filter function. You can do this ad nauseum to make steeper and steeper slopes, at the expense of worse and worse transient response and group delay. A good compromise IMHO lies somewhere around 3rd-5th order.
-Charlie
- Status
- Not open for further replies.