Right, there have been quite a few replies so I'll try to interpret / deal with the first of them - if I forget anything or get anything wrong, please feel free to correct me!
First:
I'll use L-pads to attenuate the drivers towards the lowest common denominator.
83dB (sub)woofer driver? I'll attenuate the other drivers downwards.
I'm using TPA3255 amplifier boards - 480W for the sub driver, 260W for the mid/tweeter... so it's okay to attenuate the drivers using L pads to the lowest common denominator, if I need power I'll just crank up the amp a little bit when necessary. Using L pads lowers the SNR coming out of the DSP/Amp combo.
Second:
For the crossovers in DSP, I gathered I'd use Linkwitz/Riley because:
Third:
for bass adjustment, I'm applying a Linkwitz Transform using the ADAU1701.
I've simulated this in WindISD and applied these values in SigmaStudio, inverting the values for a1 and a2 as per https://ez.analog.com/dsp/sigmadsp/f/q-a/162827/sigma-studio---linkwits-transform/413972
Result / Conclusion so far:
The current result is a speaker that actually sounds "Pretty damn okay"
So far everything has been calculated theoretically - the cabinet volume for the (sub) bass driver, the tiny volume for the BMR enclosure, the Biquads for the Linkwitz Transform to boost the bass with DSP..
So - right now all the "hardware" is set up.
Now it's "just" a matter of measuring drivers, measuring the speaker, and fooling around in software (DSP) to get the crossovers / filter right.
EDIT:
Please don't think I'm going into speaker measurements blind - I'm rolling with Vance Dickason's "LDC", Joseph D'Appolito's "Testing Loudspeakers", and Floyd Toole's "Sound Reproduction (...)"..
I have a Umik1 mic, a Dayton Dats V3 system... I will be measuring stuff ;-)
First:
I'll use L-pads to attenuate the drivers towards the lowest common denominator.
83dB (sub)woofer driver? I'll attenuate the other drivers downwards.
I'm using TPA3255 amplifier boards - 480W for the sub driver, 260W for the mid/tweeter... so it's okay to attenuate the drivers using L pads to the lowest common denominator, if I need power I'll just crank up the amp a little bit when necessary. Using L pads lowers the SNR coming out of the DSP/Amp combo.
Second:
For the crossovers in DSP, I gathered I'd use Linkwitz/Riley because:
- flat response throughout the passband
- the acoustic sum is of the drivers is unity
- zero phase difference
Third:
for bass adjustment, I'm applying a Linkwitz Transform using the ADAU1701.
I've simulated this in WindISD and applied these values in SigmaStudio, inverting the values for a1 and a2 as per https://ez.analog.com/dsp/sigmadsp/f/q-a/162827/sigma-studio---linkwits-transform/413972
Result / Conclusion so far:
The current result is a speaker that actually sounds "Pretty damn okay"
So far everything has been calculated theoretically - the cabinet volume for the (sub) bass driver, the tiny volume for the BMR enclosure, the Biquads for the Linkwitz Transform to boost the bass with DSP..
So - right now all the "hardware" is set up.
Now it's "just" a matter of measuring drivers, measuring the speaker, and fooling around in software (DSP) to get the crossovers / filter right.
EDIT:
Please don't think I'm going into speaker measurements blind - I'm rolling with Vance Dickason's "LDC", Joseph D'Appolito's "Testing Loudspeakers", and Floyd Toole's "Sound Reproduction (...)"..
I have a Umik1 mic, a Dayton Dats V3 system... I will be measuring stuff ;-)
Last edited:
LR4 filters also have the benefit of maximizing low-end driver bandwidth. It's counterintuitive but higher order filters actually increase power consumption and excursion at the highpass corner because the "knee" of the filter is so sharp and the amplitude stays flatter as the frequency falls. (This advantage of LR4 could be offset by over-exciting the driver below the box resonance if you're using a reflex port) And also there is the "ringing" ripple caused by very steep filter slopes. 4th order filters are to me the dividing line between "shallow" and "steep" and I try to stay there. They are always available in a canned DSP software package too.
If you need 8th order or higher filter slopes for your design to work consider to re-arrange the design before depending on them.
If you need more than 4 pole filters then you should use FIR complementary filters (xover) imho. Steep slope might be of help sometimes to get rid of breakup and maximising bandwidth extension.
So it tends to be of use in the mid/high range which is fine as latency is usually low up there ( even with 100db/octave and more).
In the low i do not see the advantage going over 48db/octave ( in 2 octave span the way can be totally cut which can be audible ( for a mid, let's say 250hz/1khz from some test i've done so totally subjective and it could be expectation bias from my side too...).
Pigmy, where do you want to locate your lpad and could you explain why you think they'll minimise noise in your view?
So it tends to be of use in the mid/high range which is fine as latency is usually low up there ( even with 100db/octave and more).
In the low i do not see the advantage going over 48db/octave ( in 2 octave span the way can be totally cut which can be audible ( for a mid, let's say 250hz/1khz from some test i've done so totally subjective and it could be expectation bias from my side too...).
Pigmy, where do you want to locate your lpad and could you explain why you think they'll minimise noise in your view?
The TPA3255 has >111 dB (A Weighted) SNR.
Assuming this is too noisy (don't understand why it would be) and the Lpad reduces noise by -10 dB (half as loud at 1kHz), headroom is also reduced by 10dB, the 260w amp is now the equivalent of a 26 watt amp, with a "free" room heater added.
DSP has it's best S/N when run near DBFS, so the best place to attenuate is between it and the power amp, crank down the amp gain (attenuate the input) when not needed if noise is an issue.
That assumes the drivers used have flat response throughout the passband prior to application of the DSP ;^)
Let the fun begin..
I'm using a HiVi RT1C tweeter with 94dB sensitivity.
I have the 3E Audio TPA3255 amp boards (One single channel 480W, one dual channel 260Watt) mounted right next to eachother on a piece of wood, with the DSP board and the power supply right next to them - I'm mentioning this because this might be relevant; there's no distancing nor shielding currently.
The tweeter would output noise when connected to the amp & dsp even without them having any input signal.
After putting the L pad in place, it's silent.
It's a tweeter so it's not going to need 260W, it's probably only going to need a few watts so the "now 26 watt amp" will still be overspecced.
Well yes - the fun starts with a basic working setup so I can start measuring and start making decisions based on measurements instead of manufacturers specifications
Nominal 50W is more than enough for any tweeter, for home hifi For multiway, choose best amp gain/power for each driver. If amps are 2ch and speakers 2way, you can use the high power amp to drive L and R woofers. In my 4-way speakers I have IcePower 125ASX2 for sub and woofer, and 50ASX for mid and tweeter. All drivers can still easily be pushed to overexcursion. I just don't understand modern high-power fanatism...
So I've been told. And to think that I've done it countless times for free.
Although some people do send gifts.
Although not of particular relevance to crossover design, getting rid of induced power supply noise by proper shielding, placement, design, etc. to increase signal to noise ratio is a better ultimate sonic solution than converting noise to heat after amplification of it.
Right, after another hiatus I'm back at it again..
You're right of course - the proper solution is to isolate the components to avoid interference, and in the end that will happen, that's always been the idea.
It's just that I noticed the tweeter being hissy without any audio input, in my "components screwed on a small board" test setup. Anything that helps signal to noise ratio is a win, even if later on it turns out to still be way below audible levels.
In the mean time I've replaced both my stereo amplifier (in the kitchen) and the AV receiver in the living room with alternatives that have Dirac Live (MiniDSP SHD Power for stereo, NAD T758 V3i for living room / home cinema) and I like Dirac very much, even though it's very much not a "push button" solution no matter what people say (Setting up the proper house curve and figuring out what 'curtains' to use is very personal and takes a lot of time to test).
This is relevant because now there's stuff like the MiniDSP Flex that can be used to not only do the active crossovers but also apply Dirac Live room correction at the same time.
I have all the Dayton DSP stuff figured out by now, but with the slow pace of this project I think I'll end up using something that supports Dirac Live anyway..
Moving goalposts as time goes on
Last edited:
We have a new Tech Blog which applies to this topic: Crossover Basics using miniDSP Device Console
No. The baseline is supposed to represent the needs of the cross itself. The act of flattening the "driver" is a technique whose purpose is to help those that use software which is unable to work with a target curve.. so don't read too much into it.
As the cross goes, measurements taken back from the cabinet have more relevance.. but no single response measurement is likely to be exactly correct for this purpose.
As the cross goes, measurements taken back from the cabinet have more relevance.. but no single response measurement is likely to be exactly correct for this purpose.
Near field vs far field is a question of frequency. At low frequency the driver is pistonic so near field is a good representation, and the room interaction is hard to deal with, and near field (@ low level) is a good way to deal with that (if you are much closer to the source than you are to the reflection the reflection will be small relative to the direct sound). But when the driver goes into breakup then the near field might be gathering response from a portion of the driver that doesn't represent the whole output. So you switch to far field. At higher frequencies, reflections are easier to gate out, so this works out OK, mostly. Making sure you have enough overlap to understand how one can transition into the other can be tricky.
This comes to the thing I want to say about doing crossovers in DSP: get comfortable that you understand what you see in your measurements. Its easy (even good, as mentioned earlier) to use lots of biquads to do lots of filtering. But be cautious that you're not DSP-ing measurement phantoms. Near / Far field, ground plane, windowing, in-situ, off-axis, move the speakers around, move the mic around. Try many things, see which features remain constant and which change. Depending on how much measurement experience you have, it might take some time to see which features are actual performance of the speakers, and what is an artifact of the room or the type of measurement you are performing.
This comes to the thing I want to say about doing crossovers in DSP: get comfortable that you understand what you see in your measurements. Its easy (even good, as mentioned earlier) to use lots of biquads to do lots of filtering. But be cautious that you're not DSP-ing measurement phantoms. Near / Far field, ground plane, windowing, in-situ, off-axis, move the speakers around, move the mic around. Try many things, see which features remain constant and which change. Depending on how much measurement experience you have, it might take some time to see which features are actual performance of the speakers, and what is an artifact of the room or the type of measurement you are performing.
I believe you're missing a gigantic piece of the puzzle: looking at phase (and by extension, group delay) response, in favor of only looking at only on-axis SPL (amplitude) response. Only half of the transfer function is being considered. In my experience, the phase/group delay piece of the task is probably the most important portion to get right.
Using higher order crossover filters (i.e., using the typical IIR filters found in DSP crossovers...not FIR) produces much higher phase growth distortions across the audible band and especially excess group delay spikes, which are quite audible. Not showing the phase and group delay data plots says to me that the novice DSPer is being directed not to pay attention to this aspect of the dialing-in process. For my remote dial-ins helping others to use their DSP crossovers, the phase/group delay/time alignment portion of the dial-in is the most time consuming part that requires the most attention to detail to get right.
Also, looking at phase and group delay response is one of the most direct ways of assessing whether or not the acoustic measurements are mixing strong near-field reflections with loudspeaker direct arrivals. [The problem that is the root reason why "room correction software" packages fail to get reasonable results in room--because they direct the user to place the measurement microphone much too far away from the loudspeaker under test, and are silent about dealing with acoustic floor bounce via temporarily placing extra absorption on the floor between the loudspeaker and microphone.]
Chris
Adam's instruction are good, but still miss baffle effect. In practise we must use several measurement techniques, which requires good understanding of acoustics and physics.
Delay settings must be done after individual "way" eq at eg. 1 meter on "design axis" height.
Finally "voicing" can be done at listening area/spot
Have fun when learning all this! The nicesfeature of dsp is that all this will not add any cost, just your time.
Delay settings must be done after individual "way" eq at eg. 1 meter on "design axis" height.
Finally "voicing" can be done at listening area/spot
Have fun when learning all this! The nicesfeature of dsp is that all this will not add any cost, just your time.
1st) Linearize each driver plus/minus one octave.
2nd) Apply baffle step compensation.
Does step one use near field measurements?
Hi,
Within boundaries of technique used yes (there is high frequency limit of usability). Otherwise gated measurement and if possible at different angles to be sure to correct for minimum phase and not diffraction related effects.
Bsc can only be applied second as it is room/location dependent.
Within boundaries of technique used yes (there is high frequency limit of usability). Otherwise gated measurement and if possible at different angles to be sure to correct for minimum phase and not diffraction related effects.
Bsc can only be applied second as it is room/location dependent.