35 thoughts on “What Price the Truly Rational Audio System? (About £380)”
All good stuff! I did something very similar myself using a pair of B&W 801f loudspeakers as I didn’t want to go to the trouble of making the boxes. I removed the passive crossover and used a Behringer DCX2496 to provide the crossover and three Behringer A500 amplifiers for power. I didn’t want the noise from a PC in the listening room, which is why I used the DCX rather than a PC and sound card. I then equalised the loudspeakers to be flat anechoically to +-1dB above 200Hz using tbe DEQ2496 in a combination of parametric and spot graphic EQ. With my room dimensions, I can’t measure below 200 Hz, but anyway at LF there’s a lot of room effects going on, so I’m not convinced that equalising totally flat at LF is necessary in the same way as at MF and HF.
I also completely agree with you about not equalising for a flat response in room, as that sounds unnatural, given that we are all familiar with the natural response of our rooms to everyday noises and speech. In fact, I evaluate loudspeakers by using speech rather than music, as that’s the hardest thing to get right and sounding real.
Many thanks for the comment Serge. Your lovely B&W speakers are certainly more acceptable for the living room than my speakers! I think, like me, you could never go back to passive..?
No indeed. I started with active loudspeakers in the mid 80s, with Meridian M2s, although what I would have wajted but coulkn’t afford was Meridian M10s. Then changed to Meridian DSP5000s with DSP1500 subs. I just fancied a change a couple of years ago so did the thing with the 801fs. I can’t see me changing unless it’s just wanting a change for the sake of it.
I’ll be showing some DIY omnidirectional loudspeakers at the Scalford show on the 29th March to illustrate the difference between omnis and limited directivity loudspeakers. However these are nominally full range drivers so no crossover.
Thanks for the comment. Do you mean in terms of using DSP to correct the drivers? Well yes, I think you could use DSP to flatten the phase and frequency response, but I wouldn’t particularly know how to do it. In this case, I suspect the measurements would have to be far field because there’s clever stuff going on with the whizzer cone and phase plug that will only work at a distance.
This is a really nice setup. Thanks for documenting the project. I especially like your FIR-based approach to building the x-over. Are you generating the FIR filters within your software, or using something like rePhase to generate the filters? I’m currently using a IIR-based approach with a MiniDSP to my 3-way system, but I have wanting to venture into the FIR-based crossover approach.
Thanks for the comment, Dan. I am generating the FIR filters in my software because I really wanted to be able to play with the settings in real time, and I was keen to include baffle step correction because it seems to me that this is one of the most critical aspects in getting a speaker that sounds right. I dread to think how difficult (i.e. impossible) it would be to get the same ‘finesse’ using passive crossovers only!
What size drivers do you currently use in your three way system?
Thanks for answering my questions. That makes sense to generate the FIR filters in your own program if you have the ability to write the algorithm to do it (which you do!). So the BSC is part of a single FIR filter? I.e. the woofer filer provides BSC, any EQ you are adding and also the 24 dB/octave low-pass?
I am using a 15″ woofer up to 800 Hz, a 4″ CD from 800-8kHz and a ST from 8k – 20k (passively crossed over, using the same amp that the CDs use). I have been seriously thinking of dismantling this system and building something like you have here.
But the missing piece for me is the software that will do the filtering etc.
Yes, each driver has just the one filter that combines the functions of crossover, BSC and driver correction.
It looks pretty much as though if you were using a PC this commercial package would do the job..? It includes facilities to implement the BSC. http://www.thuneau.com/allocator.htm
Is it possible to download FIR impulse responses to the MiniDSP?
Hello! This all looks very interesting and fun. I think I’ll give this a try in the summer!
Question — If I want to use my OPPO BDP-105 as a multi-channel DAC, what input can I use from my computer? Will USB handle that much data? Will it need to be HDMI? Or will I need to use a S/PDIF input? thanks!
Also, does the Thuneau software do everything your project software does?
thank you.
Hello Chris, and thanks for commenting. Just looking at the spec for your (excellent) BluRay player/DAC. It seems to suggest that it can handle 7.1 channel PCM audio via HDMI at up to 96 kHz, but only 2-channel audio via USB and S/PDIF. So you could presumably use HDMI for a three-way system (six channels).
HDMI isn’t something I’ve played with yet, but it would be a neat solution. I presume you have a multi-channel amplifier that you can feed the DAC outputs into?
As far as I can tell, the Thuneau software can do what mine can do. For the ultimate quality you will need to be able to measure the individual drivers in the near field – a port would complicate things, which is another good reason for going sealed! I use the excellent (and free) Room EQ Wizard for making the measurements. I think you could simply dial inverse-ish curves into the Thuneau software, based on the phase and frequency response plots generated by REW. On most multi-channel AV amps you seem to be able to trim the gains so you could set these to give you the flattest in-room response at the listening position when the Thuneau software is doing its stuff. REW has tools for doing this sort of thing.
Thanks for the info! That mostly made sense to me. I have a measurement microphone, mixer and Room EQ Wizard although I have not used them very much. I have a multi-channel receiver and I also have an analog multi-channel preamp. No power amps right now. I guess I’m pretty close! I’ll get reacquainted with REW and dig out an old computer. thanks again for the idea. I’ll be visiting often.
Just a thought… Most Linux installs include “MPD”, MPD (Music Player Daemon) can call external, free programs such as “Ecasound” and “Sox”. You can do FIR filters as plugins, and the result is a very nice system you can hide in a closet and control from a tablet or phone. I’ve implemented a 2-way crossover for my subs (4 by 10″) and KEF R101, and am quite happy with the results.
Thanks for the info. Yes, I can see that Linux would be an ideal way to go, except for the learning curve! – I’ll have to get round to trying to understand JACK one day…
Would the scheme you’re talking about be usable with Spotify?
Hi Andrew. That looks like a beautifully-neat system, and I would really like to know how those omnidirectional speakers sound. Got to go out right now – more later!
This is an awesome setup. I particularly like your idea of using AULab to process the audio (being a mac user myself). You could go the full “rational” by designing a FIR filter to linearize the phase in rePhase and then use a convolution reverb plug-in to convolve the input signal with the FIR filter. Your design looks similar-ish to the LXMini (with the addition of being a full 3-way, which I like better). I love your bent metal midrange/tweeter supports. Looking forward to part 2.
I’m not sure how much difference FIR/IIR or whatever phase shifting filters will actually do. I’ve listened and tried them all. So much complicated stuff happens when the sound fields are all over each other. That being said, I embarrassingly I don’t know what kind of filters are in the crossovers I’m using 🙂 Let me see if I can find them again.
Looks like a tutorial there on how to set it up too. I don’t remember this much detail when I did it 🙂 I don’t know of anyone who has done the HDMI before!
I’ve got some other things in mine – like some sample delays to centre the tweeter and so on. My design is Linkwitz inspired – this is part of Part 2. My project was two parts: 1) generate the driver signals 2) design the transducer assembly and consider all the room stuff.
This ‘blog’ is nicely written. The IIR -vs- FIR filter question is interesting. I’ve always found the lobing or waveform cancellation between drivers and L and R channels to be challenging. Try playing a 1kHz sinusoidal tone out one driver each of both channels simultaneously. Then take your phone and record the sound pressure it picks up as you walk L to R across the room in front of the speakers at 5’or 10′ away. It’s remarkable.
Well of course the room would cause the level to fluctuate wildly even from a single, mono driver, but it would all be defined by a coherent room ‘equation’ that inextricably links frequency response with time delay effects. My hand-wavey argument is to say that unless there’s a good reason to do otherwise then we want each speaker to get as close as possible to a single unitary driver that has flat phase. The room will then add its contribution, but we are very good at hearing past room acoustics, and don’t normally notice it. We may not be so good at hearing past complex, arbitrary interactions between multiple drivers, which is much less of a ‘natural’ effect, I would guess.
In some of my previous posts I point to reviews of DSP-based speakers with phase correction where the reviewers do seem to be hearing something special. But I accept it could be that they imagined it, or it was merely active three-way-ness that was impressing them.
I certainly wouldn’t like to bet that I could discern between flat phase response and non-flat on any particular piece of music or test signal, but then that isn’t really an argument against aspiring to a flat phase response on ‘rational’ grounds. Indeed, if it costs nothing to implement I will implement it anyway and then know that that particular box is ticked and not worry about it again! The conventional argument against this seems to be a bit backwards i.e. that because it is not possible to achieve flat phase response using standard audiophile hardware at any price level, then it cannot matter.
If there was another factor, like flat phase response could only be achieved at the cost of higher latency, rendering the system’s use with video impractical, then I might have to think again. But on the other hand, the rational response might be to introduce a simple video delay line to sync the video with the delayed audio…
“Well of course the room would cause the level to fluctuate wildly even from a single, mono driver, but it would all be defined by a coherent room ‘equation’ that inextricably links frequency response with time delay effects.”
I don’t think you can counter the room response in any kind of practical way except for timing the arrival of surround sound drivers and maybe getting rid of low frequency nodes. I have tried. Anyway, that’s not the point. There is a huge difference between walking in front of a mono driver and two L,R drivers playing the same tone. It made me rethink everyting with respect to phase. It happens independent of room reflections. It’s all to do with the direct radiated sound. That’s why I said to try it. I’ll make a short test video to illustrate.
“My hand-wavey argument is to say that unless there’s a good reason to do otherwise then we want each speaker to get as close as possible to a single unitary driver that has flat phase.”
Good rational (but not scientific) argument – but then making the M and H drivers point source is also important. This is included in my (Linkwitz’s) design. But in your design you get inherent lobing between the M and H drivers even if you have a flat phase crossover. Anyway, I’m not sure it’s even that relevant in our applications – especially when compared to the inherent L,R interference patterns and phase shift involved when you move your head a few cm in the sound field as shown in the video.. And I have good reason to not use FIR filters. I don’t have FIR Audio Units.
“between multiple drivers, which is much less of a ‘natural’ effect, I would guess.”
To say that we can filter the boxy room reflections but not filter the multiple driver effects is not very scientific – or even rational for that matter. Why one and not the other? Anyway, we seem to filter what I show in the video and that’s not a subtle effect.
“In some of my previous posts I point to reviews of DSP-based speakers with phase correction where the reviewers do seem to be hearing something special.”
All audiophiles claim to be hearing something special about something they have faith in. lolz.
“But I accept it could be that they imagined it, or it was merely active three-way-ness that was impressing them.”
The latter would be my guess. Three way actives kick serious butt. Seriously.
“but then that isn’t really an argument against aspiring to a flat phase response on ‘rational’ grounds.”
You beat me to that obvious point 🙂
“Indeed, if it costs nothing to implement I will implement it anyway and then know that that particular box is ticked and not worry about it again!”
Logical. But you might want to take linear phase out of your (otherwise awesome) Signal Path of Virtue since it’s not a rational argument. As you said. For example, it would be a shame if someone wanted to implement my approach but then felt they were hindered by suboptimal digital IIR crossover filters because it was not aligned (NPI) with your proposed signal path of virtue. Especially when my approach may, in fact, be a more virtuous path; both cheaper and easier and without caveats.
“The conventional argument against this seems to be a bit backwards i.e. that because it is not possible to achieve flat phase response using standard audiophile hardware at any price level, then it cannot matter.”
Fair point. But there has also been decades of work in this prior area. For me, it matters if you can create an experiment demonstrate that it matters. I have tried. I have failed to obtain conclusive results on this topic. So I don’t worry about my IIRs for the moment. Less worry = more listening pleasure in my experience. Time alignment also has less effect than I would have imagined…. I try changing it around sometimes for kicks. It’s incredible what our brain keeps and what it filters out.
“But on the other hand, the rational response might be to introduce a simple video delay line to sync the video with the delayed audio…”
Again, beat me to my question 🙂 What is the latency of your FIR filters? How long are they? I remember worrying about audio delay wrt video when I did my implementation. Turned out not to be a problem. My DSP crossover uses less than 5% CPU. I’m not sure how long it takes.
“I don’t think you can counter the room response in any kind of practical way…”. I agree, and I think I say somewhere that I don’t worry about it unduly, making the point that a real musician could play in any room, and at no point would it sound wrong per se, even though you could show that the room is having huge effects on the frequency response compared to an anechoic chamber. This is the thing about our ability to “hear past” the room’s acoustics which is something we seem to do quite naturally. A demonstration of this is to make a recording of someone talking in a room from several feet away. While their voice sounds perfectly normal when you are there, once out of the context of the room and listening to the recording, the room’s acoustics are laid bare.
As a statement of fact, the effects of the room are ‘static’ (i.e. the room isn’t changing shape dynamically), and are linear and ‘consistent’ i.e. for every fluctuation in frequency response there is a corresponding delayed reflection which our ears pick up as delays in the time domain, not just the frequency magnitude. The whole thing hangs together as we turn our heads or move around the room. And of course the direct sounds from the performer to our ears are not affected by the room. I would contrast this with splitting the audio up into several frequency bands and applying arbitrary delays and phase shifts to each band. Does anything like this ever happen in nature? Is there any reason to think that our ears and brains have evolved to ignore, or cope with this? Does it, in fact, resemble an acoustic that could never exist, or a room whose shape is changing dynamically? Maybe we can’t consciously discern it, but is it a fatiguing effect nevertheless?
I know that there has been work done on the audibility of phase in the past, but I would ask the following:
(a) Was the audibility of ‘phase shift’ vs. ‘no phase shift’ tested? From what I have seen, it was usually ‘phase shift’ vs. ‘different phase shift’.
(b) What sort of music and recordings was it tried with? I think that this choice probably limits the usefulness of all listening tests (see my post on Science and Audio for my scepticism regarding listening tests!).
Here’s a link to an extremely interesting speaker review where the author tells us how messed-up phase unambiguously “colours” the sound of transients, even though the steady state frequency response can be perfectly flat: http://www.iar-80.com/page96.html
Another one of my pet subjects is bass reflex speakers. Can we state that we know what effect the room has on the bass if the speaker itself is doing strange, unnatural things with the bass? Are audiophiles spending inordinate amounts of time and money trying to fix the room with treatments and/or EQ when the “problems” are being compounded by an additional (unnecessary?) resonator?
I realise that my speakers are not a point source, but I did try to get the drivers as close together as possible. As I say in the article above, there are people who argue that speakers should be wide flat-fronted boxes, and others who say they should be omnidirectional. Mine are sort of in between, so they might be the perfect compromise, or the worst of all worlds! They seem pretty good to me, anyway.
Turning the question of flat phase response around: given that I can program in any phase response I want, what should that response be? I would think it obvious that I want the acoustic phase of the drivers to overlap perfectly as they cross over. But beyond that, do I want the overall response of the speaker to be flat, or ‘minimum phase’ or what? To me, the only ‘natural’ response is the flat (zero) phase (plus an inevitable delay). Other questions might override that decision, such as latency. In my system the latency is huge: about a second, so totally unsuitable for video work unless the video can be delayed too. I notice that other FIR-based linear phase systems are similarly burdened. The latency can be reduced, with the accuracy of the filters gradually being compromised, but it’s not something I have tried – yet. If the video can be delayed then I don’t think it’s an issue (except for the system’s responsiveness to the controls being operated..?)
“I would contrast this with splitting the audio up into several frequency bands and applying arbitrary delays and phase shifts to each band. Does anything like this ever happen in nature? Is there any reason to think that our ears and brains have evolved to ignore, or cope with this? Does it, in fact, resemble an acoustic that could never exist, or a room whose shape is changing dynamically? Maybe we can’t consciously discern it, but is it a fatiguing effect nevertheless?”
Very fair points and questions in my opinion. I’m not sure two point sources generating the information for a ‘centred’ audio images also happens in nature. That 1kHz test I will show is extremely ‘fatiguing’ to listen to. And by definition attempting to create a natural sound field with a L and R channel located anywhere other than directly over your ears yields an unnatural result in general. Binaural recordings seem to be the only way to reproduce the accurate sounds fields that your brain finds no fault with – in my experience at least.
“there has been work done on the audibility of phase in the past, but I would ask the following:
(a) Was the audibility of ‘phase shift’ vs. ‘no phase shift’ tested? From what I have seen, it was usually ‘phase shift’ vs. ‘different phase shift’.
(b) What sort of music and recordings was it tried with? I think that this choice probably limits the usefulness of all listening tests (see my post on Science and Audio for my scepticism regarding listening tests!).”
Very good questions. We both have the equipment to test and generate insight. Btw I notice that you blog has lots in the way of ‘thinking’ but it’s light on talking about your listening experiences and experimental investigations. The listening exploration would be very interesting. Especially when based on rational curiosities!
“so they might be the perfect compromise, or the worst of all worlds! They seem pretty good to me, anyway”
I’m not surprised at all that they sound fantastic. You hit all the main considerations and the remainder may not be that important. ‘Maybe’ side by side comparison would yield different results between designs – but what would be the point? You probably have more than what you need for a very pleasurable and visceral experience of an otherwise flawed two channel audio scene reproduction technology. That being said, the L-R technology can still be fantastic to listen to.
BTW I took my design don’t to *THE* hifi store in town and successfully put it up against some very expensive hardware. I rolled in with the amp in my pack back and a speaker in each hand. Hilarious. Mine didn’t sound conclusively *better* but for 500$ in parts there’s no way any rational person would spend even twice as much for the other stuff – unless they like ‘art’ pieces. Which they seem to.
“Turning the question of flat phase response around: given that I can program in any phase response I want, what should that response be? I would think it obvious that I want the acoustic phase of the drivers to overlap perfectly as they cross over. But beyond that, do I want the overall response of the speaker to be flat, or ‘minimum phase’ or what? To me, the only ‘natural’ response is the flat (zero) phase (plus an inevitable delay).”
Maybe it just doesn’t matter? What’s next, controlling room reflections (vertical flat services also being unnatural)? If anything, I think the next step is experimenting with the all the great questions above. Try stuff and see. I don’t think you can think your way out of it from here.
“Other questions might override that decision, such as latency. In my system the latency is huge: about a second, so totally unsuitable for video work unless the video can be delayed too. I notice that other FIR-based linear phase systems are similarly burdened.”
Well, now we can see the inherent FIR problem. My classic IIR DSP filters are just a handful of coefficients and very fast to calculate. I use my system for video. And generally I (and other users) would find a 1s latency to be intolerable when operating controls. It’s bad enough I have the ugly speakers in the living room. lolol. I don’t need any other inconveniences that may jeopardize the somewhat invasive project in my household.
However, I am still very interested in answers to your questions.
Maybe you can grab an OSX HDMI-enabled device and a cheap 7.1 HDMI AVR and try my IIR approach for comparison? Maybe you can run a Hackintosh? You may even appreciate the lack of wires and elegance in general. Not to mention that you get the full OSX video experience for free too. Just plug in a monitor. Or a projector.
Also, FWIW I have listened to many fantastic sounding hifi systems in recent years. Some DSP based. Some based on first order passive crossover filters. None of this crossover stuff is obvious except for the purely rational stuff we can think through. Like how I can generate 1200WPC (peak) of clean audio amplification from a raw digital signal using a 200$ commodity amplifier. Awesome. Can’t do that with a three way passive approach.
This is what I have implemented now in my system. I am using the very excellent and cheap (purchased mine for $55 off of eBay) Pioneer VSX-D912 receiver. My system uses IIR filters (by way of a MiniDSP 4x10HD). Tri-amped JBL speakers. 15″ 2226, 2″ compression driver (2445) and super tweeter (2405). Most of my listening is done at -40 dB on the receiver (lots of headroom to spare!).
I do like the idea of experimenting with the Mac-based system, though. Seems like a lot more flexibility to try and compare IIR/FIR-based crossover filters, linear/minimum phase etc.
@Andy, what Sony receiver do you use? My Pioneer does not have HDMI in it (hence the low price) but it is 100 watts/channel and has plenty of power to spare (especially with the high-efficiency horns).
” I’m not sure two point sources generating the information for a ‘centred’ audio images also happens in nature. ”
No, I suppose not, but it is wrong in quite a ‘simple’ way! If there are supposedly-horrendous things happening with phase shifts etc. as you move around the room then they’re not obvious as you listen. And the fact that both ears are receiving both speakers’ output doesn’t seem to stop you from getting a pretty good illusion of a “soundstage”. Each speaker may be individually correct, so in itself the result is two ‘correct’ speakers playing in unison within your fixed, static and consistent room acoustics. As I say, without proof of any sort, I imagine that this will be easier to listen to than speakers that are applying arbitrary frequency-dependent phase shifts. But I do realise that very few people in the audiophile world give this argument any credence at all. It makes me happy, anyway.
I recently listened to some ambiophonic recording examples (you have to push your speakers closer together) and they were pretty good. A deliberate attempt to make speaker-based stereo more ‘correct’ and, I suppose, ‘natural’. However, the algorithm to do this involves each speaker recursively producing attenuated, delayed, inverted versions of the other’s output signal. Refinements include limiting this to the mid-range. As I say it sounds good, but on the examples I heard it came as a relief to eventually go back to plain stereo. Maybe if I had a working system to do the correction myself I could find better settings.
Did you see that speaker review I linked to? It’s an amazing piece of work, and talks about something that I have rarely seen mentioned: the idea that unless all frequency components are lined up with the correct timing and phase, it will “colour” the sound of transients. By going with linear phase filters I can cross that potential problem off the list without doing any experiments.
“I notice that you blog has lots in the way of ‘thinking’ but it’s light on talking about your listening experiences and experimental investigations…”
Yes, that’s the “rational” aspect. I think we could spend the rest of our lives experimenting in semi-random fashion but not come to any conclusions. I am trying to do it the other way round: find the most blameless solution I can think of that is sufficiently practical for my purposes, implement it, and cross it off the list. It frees me up to then concentrate on the interesting stuff such as maybe putting together some omnidirectional speakers like yours…
“No, I suppose not, but it is wrong in quite a ‘simple’ way! If there are supposedly-horrendous things happening with phase shifts etc. as you move around the room then they’re not obvious as you listen.”
There are horrendous things happening. And your brain seems to deal with when listening to full spectrum sound with a L-R channel stereo system. It’s really quite amazing.
The video below shows full 0-360 deg phase interference from a supposed stereo image located between the two transducers. And this is with room reflections. It would be much more pronounced without the room influence. Higher frequencies result in closer spacing.
Anyway, all that above makes me think that the crossover phase question may not be that relevant in the scheme of things. But I have no real conclusive arguments to support that.
“And the fact that both ears are receiving both speakers’ output doesn’t seem to stop you from getting a pretty good illusion of a “soundstage”.”
Yup, this is true. Lucky for us. We seem to filter *a lot*. Next time I’m told that I have selective hearing……
“Each speaker may be individually correct, so in itself the result is two ‘correct’ speakers playing in unison within your fixed, static and consistent room acoustics. ”
It’s still my observation that two ‘correct’ speakers in a room trying to create a centered image is an inherently flawed approach regardless of how static and consistent the environment is.
“I recently listened to some ambiophonic recording examples (you have to push your speakers closer together) and they were pretty good. A deliberate attempt to make speaker-based stereo more ‘correct’ and, I suppose, ‘natural’. However, the algorithm to do this involves each speaker recursively producing attenuated, delayed, inverted versions of the other’s output signal. ”
I know this research. I don’t really see the point for Hi-Fi. That binaural link I posted really takes care of everything in the best possible way – except for the sound you feel with your body and not your eardrums. My impression is that the researcher would like to get it into 3DTV applications where the lack of fidelity would be less of an issue. Or perhaps for iPad or laptop speakers. All huge revenue opportunities. hmmmm maybe we picked the wrong projects!
On the other hand I’ve experimented with introducing ‘cross talk’ for headphone listening. The process where you ‘place’ the speakers in front of you by strategically cross-feeding the signal. That’s really great an eliminates the centred image problems discussed above.
“Did you see that speaker review I linked to? It’s an amazing piece of work, and talks about something that I have rarely seen mentioned: the idea that unless all frequency components are lined up with the correct timing and phase, it will “colour” the sound of transients. By going with linear phase filters I can cross that potential problem off the list without doing any experiments.”
This is generally a slippery slope of acceptance. In the same way I could blindly use gold plated power cords so I don’t have to worry about how the copper oxidization influences the resultant sound. I suggest that generally there is more of a price to pay for FIR filters than IIR so it’s worth some investigation of their pros and cons. That being said, I don’t think we’ll find anything conclusive to make one approach more compelling that the other in any kind of audible way – especially given the above. But you never know.
I’m not sure about that article. It seems written from an emotional place and has all the hallmarks of audiophile kool-aid. I get what’s being said but there must be a better way to evaluate. Besides the listener really has no way to know what he is listening to. He thinks it might be phase colouration. Perhaps it’s something completely different. Perhaps it’s nothing? No real way to know. I drink plenty of kool-aid myself from time to time tho. It tastes good going down.
“Yes, that’s the “rational” aspect. I think we could spend the rest of our lives experimenting in semi-random fashion but not come to any conclusions.”
Who said anything about semi-random? And experiments don’t necessarily have to be time consuming. It’s just more about looking for answers in the physical world rather than on the internet.
“Find the most blameless solution I can think of that is sufficiently practical for my purposes, implement it, and cross it off the list.”
This is really the best approach. Are the FIR filters for you any more practical than the more efficient IIR filters?
On an unrelated characteristic of the Omni design: the two way version can have outstanding bass if you don’t want to listen too loud. The travel of the mid-woofer is fantastic and the addition of some digital bass lift really produces dramatic results. Again, this would be highly impractical with a passive crossover.
I have about 10 more of those bass tubes sitting around since I was forced to buy an entire 20′ green sewer pipe. I can send you (anyone) some if you (they) pay for shipping 🙂
“This is generally a slippery slope of acceptance. In the same way I could blindly use gold plated power cords so I don’t have to worry about how the copper oxidization influences the resultant sound. I suggest that generally there is more of a price to pay for FIR filters than IIR so it’s worth some investigation of their pros and cons. That being said, I don’t think we’ll find anything conclusive to make one approach more compelling that the other in any kind of audible way – especially given the above. But you never know.”
The difference being that if we tried to measure the effect of the gold plated power cord we would not detect anything at all because its effects would be less than 1 LSB of a 20 bit signal – probably, if I had to bet on it. But the phase thing is unambiguously measurable. The question is whether is it audible or not. Maybe it is not audible 99% of the time. Or not audible at all on certain types of music.
There is one listening test where I can always identify the difference between ‘correct’ and ‘otherwise’ when switching between the two. If I play noise through the system and turn off the correction I can hear the effect. This isn’t just a test of overall phase correctness, however. This is the difference between a suckout and no suckout because the correction is affecting the relative phase between the drivers as well as the overall phase linearity. Applying the correction, it is clear that a notch in the response is being ‘filled out’. Similarly, adjusting the relative delay between the drivers it is clear when they are best aligned at the listening position and therefore cancelling each other out the least. With certain types of music this is far from clearly audible. It seems to be most obvious on music which features a ‘wash’ of overdriven electric guitar. But my anecdotal subjective experience is that after a while without the correction, there does seem to be an edginess or hardness to the sound regardless of the type of music. Such a statement is not “scientific”, though, which is why I simply use the corrected FIR filters and forget about it. The latency of the large FIR filters I have chosen to use doesn’t bother me. Two years in, and happily living with a second’s latency – it is a lot less than the latency of vinyl!
“The difference being that if we tried to measure the effect of the gold plated power cord we would not detect anything at all because its effects would be less than 1 LSB of a 20 bit signal – probably, if I had to bet on it. But the phase thing is unambiguously measurable. The question is whether is it audible or not. Maybe it is not audible 99% of the time. Or not audible at all on certain types of music.”
Fair argument. However, I can measure the increase in oxidation of the plug contacts to a much more significant effect than that. Also, I can measure the difference between the inductive and capacitive nature various cords as well. Is it audible? I know what I think about that…
“There is one listening test where I can always identify the difference between ‘correct’ and ‘otherwise’ when switching between the two. If I play noise through the system and turn off the correction I can hear the effect.”
If you play noise and switch between *anything* you will hear a difference. I do this with my time corrections and various other filters etc. A difference between one or the other is very clear with noise – much more clear than with music. It’s probably subjective as to if it’s filling a ‘notch’ or not; that’s a qualitative assessment. Playing music and assessing which is more accurate or images better or sound less smudged is very difficult for me; even if I take settings way out of where they *should* be to make rational sense. The mind is a funny thing.
This afternoon I started a new room assessment technique. A 10Hz audio square wave generator and a listening microphone fed into a scope. So far so good. Forget the freq-domain stuff 🙂 I’ve never looked at my system like this before in the time domain. I don’t know why.
“while without the correction, there does seem to be an edginess or hardness to the sound regardless of the type of music.”
Interesting. Wonder if it’d be the same with IIR filters?
“I can measure the increase in oxidation of the plug contacts to a much more significant effect than that. Also, I can measure the difference between the inductive and capacitive nature various cords as well.”
Of course we can measure these, but do they actually have any measurable effect on the audio? I have noticed that cable enthusiasts (not implying that you are) often refer to the widely-varying parameters of L, C and R between different cables, but don’t take it to its conclusion and show how these would affect things in the context of the circuit. They fail to show that increasing the resistance of a signal cable by a factor of a thousand might reduce the volume by 0.0001 dB, or whatever, and that its effects are predictable and benign. In terms of oxidation, if the power cord contacts are ‘wiped’ every time the plug is inserted, does the oxidation actually have any effect at all?
“It’s probably subjective as to if it’s filling a ‘notch’ or not; that’s a qualitative assessment. ”
I am not saying that I can listen to a single sample of noise and say whether the response is flat or not or in what way it is not flat, but I can clearly hear the difference when the notch (that in this case I know is there) is filled in – and comparing the two conditions side by side I could tell you which was which.
“Interesting. Wonder if it’d be the same with IIR filters?”
Well the implication is that if you measure your drivers’ raw phase responses and they are mismatched through the crossover, then even if you use crossover filters with perfectly-matched phase you will still get an acoustic suck-out. The benefit of measuring the system from end to end is that you automatically compensate for all the sources of phase shift and, in conjunction with the adjustable delay, you can set the system to something close to optimum. You can always say “Ah, but is it optimum? Can you prove that suck-outs aren’t beneficial to the sound?” etc. but in this case I am happy to go for the straightforwardly-obvious setup.
“Because you have to change records? ”
Of course. And get the LP out of its sleeve, carefully place it on the platter, cue up the arm, wait while it drops onto the run-in, and wait for a few seconds for the music to start. The “latency” and lack of responsiveness of the controls is huge!
I’m not a cable enthusiast. I was just making the point that one can ‘cast doubt’ of throw irrelevant monkeys in to any technical argument. Very good points wrt the FIR approach. I may have to revisit this. Thanks.
What do we think about something like the LX Mini by SL? Seems to share the same mid-bass driver as @Andy’s omni loudspeaker system. The woofer is in a sealed enclosure, it is bi-amped, the system uses DSPs and SL recommends a sensible 4-channel solid state amp. It’s a 2-way, not a 3-way but other than that, it seems pretty sensible to me (and priced quite nicely!). I’ve been on the fence about building that setup, or a system more similar to what you have (3-way sealed boxes with a large LF driver).
My design is a modified SL Pluto – the predecessor of the mini. I’ll modify mine to the LX Mini tweeter in due time. As I mentioned the LX Mini bass is surprisingly impressive as a bass lifted two way. It’s probably fine for most rooms unless you like it quite loud. I added the 10″ woofers because a) I like it loud, b) i like experimenting (I can compare the two), c) I had 7 amp channels anyway, d) I needed a wider base so it would stand up properly.
The Rational Audiophile 
All good stuff! I did something very similar myself using a pair of B&W 801f loudspeakers as I didn’t want to go to the trouble of making the boxes. I removed the passive crossover and used a Behringer DCX2496 to provide the crossover and three Behringer A500 amplifiers for power. I didn’t want the noise from a PC in the listening room, which is why I used the DCX rather than a PC and sound card. I then equalised the loudspeakers to be flat anechoically to +-1dB above 200Hz using tbe DEQ2496 in a combination of parametric and spot graphic EQ. With my room dimensions, I can’t measure below 200 Hz, but anyway at LF there’s a lot of room effects going on, so I’m not convinced that equalising totally flat at LF is necessary in the same way as at MF and HF.
I also completely agree with you about not equalising for a flat response in room, as that sounds unnatural, given that we are all familiar with the natural response of our rooms to everyday noises and speech. In fact, I evaluate loudspeakers by using speech rather than music, as that’s the hardest thing to get right and sounding real.
Measurements and a description of my efforts on https://sites.google.com/site/audiopages/
Best regards
Serge
LikeLike
Many thanks for the comment Serge. Your lovely B&W speakers are certainly more acceptable for the living room than my speakers! I think, like me, you could never go back to passive..?
LikeLike
No indeed. I started with active loudspeakers in the mid 80s, with Meridian M2s, although what I would have wajted but coulkn’t afford was Meridian M10s. Then changed to Meridian DSP5000s with DSP1500 subs. I just fancied a change a couple of years ago so did the thing with the 801fs. I can’t see me changing unless it’s just wanting a change for the sake of it.
I’ll be showing some DIY omnidirectional loudspeakers at the Scalford show on the 29th March to illustrate the difference between omnis and limited directivity loudspeakers. However these are nominally full range drivers so no crossover.
S
LikeLike
Do you think this would work with an single driver speaker like those from Audio Nirvana?
LikeLike
Thanks for the comment. Do you mean in terms of using DSP to correct the drivers? Well yes, I think you could use DSP to flatten the phase and frequency response, but I wouldn’t particularly know how to do it. In this case, I suspect the measurements would have to be far field because there’s clever stuff going on with the whizzer cone and phase plug that will only work at a distance.
LikeLike
This is a really nice setup. Thanks for documenting the project. I especially like your FIR-based approach to building the x-over. Are you generating the FIR filters within your software, or using something like rePhase to generate the filters? I’m currently using a IIR-based approach with a MiniDSP to my 3-way system, but I have wanting to venture into the FIR-based crossover approach.
LikeLike
Thanks for the comment, Dan. I am generating the FIR filters in my software because I really wanted to be able to play with the settings in real time, and I was keen to include baffle step correction because it seems to me that this is one of the most critical aspects in getting a speaker that sounds right. I dread to think how difficult (i.e. impossible) it would be to get the same ‘finesse’ using passive crossovers only!
What size drivers do you currently use in your three way system?
LikeLike
Thanks for answering my questions. That makes sense to generate the FIR filters in your own program if you have the ability to write the algorithm to do it (which you do!). So the BSC is part of a single FIR filter? I.e. the woofer filer provides BSC, any EQ you are adding and also the 24 dB/octave low-pass?
BTW here is my system:
http://www.audiokarma.org/forums/showthread.php?t=609742
I am using a 15″ woofer up to 800 Hz, a 4″ CD from 800-8kHz and a ST from 8k – 20k (passively crossed over, using the same amp that the CDs use). I have been seriously thinking of dismantling this system and building something like you have here.
But the missing piece for me is the software that will do the filtering etc.
LikeLike
Yes, each driver has just the one filter that combines the functions of crossover, BSC and driver correction.
It looks pretty much as though if you were using a PC this commercial package would do the job..? It includes facilities to implement the BSC.
http://www.thuneau.com/allocator.htm
Is it possible to download FIR impulse responses to the MiniDSP?
LikeLike
Hello! This all looks very interesting and fun. I think I’ll give this a try in the summer!
Question — If I want to use my OPPO BDP-105 as a multi-channel DAC, what input can I use from my computer? Will USB handle that much data? Will it need to be HDMI? Or will I need to use a S/PDIF input? thanks!
Also, does the Thuneau software do everything your project software does?
thank you.
LikeLike
Hello Chris, and thanks for commenting. Just looking at the spec for your (excellent) BluRay player/DAC. It seems to suggest that it can handle 7.1 channel PCM audio via HDMI at up to 96 kHz, but only 2-channel audio via USB and S/PDIF. So you could presumably use HDMI for a three-way system (six channels).
HDMI isn’t something I’ve played with yet, but it would be a neat solution. I presume you have a multi-channel amplifier that you can feed the DAC outputs into?
As far as I can tell, the Thuneau software can do what mine can do. For the ultimate quality you will need to be able to measure the individual drivers in the near field – a port would complicate things, which is another good reason for going sealed! I use the excellent (and free) Room EQ Wizard for making the measurements. I think you could simply dial inverse-ish curves into the Thuneau software, based on the phase and frequency response plots generated by REW. On most multi-channel AV amps you seem to be able to trim the gains so you could set these to give you the flattest in-room response at the listening position when the Thuneau software is doing its stuff. REW has tools for doing this sort of thing.
LikeLike
Thanks for the info! That mostly made sense to me. I have a measurement microphone, mixer and Room EQ Wizard although I have not used them very much. I have a multi-channel receiver and I also have an analog multi-channel preamp. No power amps right now. I guess I’m pretty close! I’ll get reacquainted with REW and dig out an old computer. thanks again for the idea. I’ll be visiting often.
LikeLike
Just a thought… Most Linux installs include “MPD”, MPD (Music Player Daemon) can call external, free programs such as “Ecasound” and “Sox”. You can do FIR filters as plugins, and the result is a very nice system you can hide in a closet and control from a tablet or phone. I’ve implemented a 2-way crossover for my subs (4 by 10″) and KEF R101, and am quite happy with the results.
LikeLiked by 1 person
Thanks for the info. Yes, I can see that Linux would be an ideal way to go, except for the learning curve! – I’ll have to get round to trying to understand JACK one day…
Would the scheme you’re talking about be usable with Spotify?
LikeLike
I just did the following years ago. Easy and low-cost. No performance caveats to speak of.
LikeLiked by 1 person
Hi Andrew. That looks like a beautifully-neat system, and I would really like to know how those omnidirectional speakers sound. Got to go out right now – more later!
LikeLike
This is an awesome setup. I particularly like your idea of using AULab to process the audio (being a mac user myself). You could go the full “rational” by designing a FIR filter to linearize the phase in rePhase and then use a convolution reverb plug-in to convolve the input signal with the FIR filter. Your design looks similar-ish to the LXMini (with the addition of being a full 3-way, which I like better). I love your bent metal midrange/tweeter supports. Looking forward to part 2.
LikeLike
I’m not sure how much difference FIR/IIR or whatever phase shifting filters will actually do. I’ve listened and tried them all. So much complicated stuff happens when the sound fields are all over each other. That being said, I embarrassingly I don’t know what kind of filters are in the crossovers I’m using 🙂 Let me see if I can find them again.
Here:
http://www.eecs.qmul.ac.uk/~andrewm/composer/tech.html#crossoverUnit
Looks like a tutorial there on how to set it up too. I don’t remember this much detail when I did it 🙂 I don’t know of anyone who has done the HDMI before!
I’ve got some other things in mine – like some sample delays to centre the tweeter and so on. My design is Linkwitz inspired – this is part of Part 2. My project was two parts: 1) generate the driver signals 2) design the transducer assembly and consider all the room stuff.
LikeLike
Yeah, it’s fun. Good for streaming audio and great for nexflix!!
LikeLike
This ‘blog’ is nicely written. The IIR -vs- FIR filter question is interesting. I’ve always found the lobing or waveform cancellation between drivers and L and R channels to be challenging. Try playing a 1kHz sinusoidal tone out one driver each of both channels simultaneously. Then take your phone and record the sound pressure it picks up as you walk L to R across the room in front of the speakers at 5’or 10′ away. It’s remarkable.
LikeLike
Thanks for the comment.
Well of course the room would cause the level to fluctuate wildly even from a single, mono driver, but it would all be defined by a coherent room ‘equation’ that inextricably links frequency response with time delay effects. My hand-wavey argument is to say that unless there’s a good reason to do otherwise then we want each speaker to get as close as possible to a single unitary driver that has flat phase. The room will then add its contribution, but we are very good at hearing past room acoustics, and don’t normally notice it. We may not be so good at hearing past complex, arbitrary interactions between multiple drivers, which is much less of a ‘natural’ effect, I would guess.
In some of my previous posts I point to reviews of DSP-based speakers with phase correction where the reviewers do seem to be hearing something special. But I accept it could be that they imagined it, or it was merely active three-way-ness that was impressing them.
I certainly wouldn’t like to bet that I could discern between flat phase response and non-flat on any particular piece of music or test signal, but then that isn’t really an argument against aspiring to a flat phase response on ‘rational’ grounds. Indeed, if it costs nothing to implement I will implement it anyway and then know that that particular box is ticked and not worry about it again! The conventional argument against this seems to be a bit backwards i.e. that because it is not possible to achieve flat phase response using standard audiophile hardware at any price level, then it cannot matter.
If there was another factor, like flat phase response could only be achieved at the cost of higher latency, rendering the system’s use with video impractical, then I might have to think again. But on the other hand, the rational response might be to introduce a simple video delay line to sync the video with the delayed audio…
LikeLike
Great response! Thanks.
“Well of course the room would cause the level to fluctuate wildly even from a single, mono driver, but it would all be defined by a coherent room ‘equation’ that inextricably links frequency response with time delay effects.”
I don’t think you can counter the room response in any kind of practical way except for timing the arrival of surround sound drivers and maybe getting rid of low frequency nodes. I have tried. Anyway, that’s not the point. There is a huge difference between walking in front of a mono driver and two L,R drivers playing the same tone. It made me rethink everyting with respect to phase. It happens independent of room reflections. It’s all to do with the direct radiated sound. That’s why I said to try it. I’ll make a short test video to illustrate.
“My hand-wavey argument is to say that unless there’s a good reason to do otherwise then we want each speaker to get as close as possible to a single unitary driver that has flat phase.”
Good rational (but not scientific) argument – but then making the M and H drivers point source is also important. This is included in my (Linkwitz’s) design. But in your design you get inherent lobing between the M and H drivers even if you have a flat phase crossover. Anyway, I’m not sure it’s even that relevant in our applications – especially when compared to the inherent L,R interference patterns and phase shift involved when you move your head a few cm in the sound field as shown in the video.. And I have good reason to not use FIR filters. I don’t have FIR Audio Units.
“between multiple drivers, which is much less of a ‘natural’ effect, I would guess.”
To say that we can filter the boxy room reflections but not filter the multiple driver effects is not very scientific – or even rational for that matter. Why one and not the other? Anyway, we seem to filter what I show in the video and that’s not a subtle effect.
“In some of my previous posts I point to reviews of DSP-based speakers with phase correction where the reviewers do seem to be hearing something special.”
All audiophiles claim to be hearing something special about something they have faith in. lolz.
“But I accept it could be that they imagined it, or it was merely active three-way-ness that was impressing them.”
The latter would be my guess. Three way actives kick serious butt. Seriously.
“but then that isn’t really an argument against aspiring to a flat phase response on ‘rational’ grounds.”
You beat me to that obvious point 🙂
“Indeed, if it costs nothing to implement I will implement it anyway and then know that that particular box is ticked and not worry about it again!”
Logical. But you might want to take linear phase out of your (otherwise awesome) Signal Path of Virtue since it’s not a rational argument. As you said. For example, it would be a shame if someone wanted to implement my approach but then felt they were hindered by suboptimal digital IIR crossover filters because it was not aligned (NPI) with your proposed signal path of virtue. Especially when my approach may, in fact, be a more virtuous path; both cheaper and easier and without caveats.
“The conventional argument against this seems to be a bit backwards i.e. that because it is not possible to achieve flat phase response using standard audiophile hardware at any price level, then it cannot matter.”
Fair point. But there has also been decades of work in this prior area. For me, it matters if you can create an experiment demonstrate that it matters. I have tried. I have failed to obtain conclusive results on this topic. So I don’t worry about my IIRs for the moment. Less worry = more listening pleasure in my experience. Time alignment also has less effect than I would have imagined…. I try changing it around sometimes for kicks. It’s incredible what our brain keeps and what it filters out.
“But on the other hand, the rational response might be to introduce a simple video delay line to sync the video with the delayed audio…”
Again, beat me to my question 🙂 What is the latency of your FIR filters? How long are they? I remember worrying about audio delay wrt video when I did my implementation. Turned out not to be a problem. My DSP crossover uses less than 5% CPU. I’m not sure how long it takes.
A
LikeLike
“I don’t think you can counter the room response in any kind of practical way…”. I agree, and I think I say somewhere that I don’t worry about it unduly, making the point that a real musician could play in any room, and at no point would it sound wrong per se, even though you could show that the room is having huge effects on the frequency response compared to an anechoic chamber. This is the thing about our ability to “hear past” the room’s acoustics which is something we seem to do quite naturally. A demonstration of this is to make a recording of someone talking in a room from several feet away. While their voice sounds perfectly normal when you are there, once out of the context of the room and listening to the recording, the room’s acoustics are laid bare.
As a statement of fact, the effects of the room are ‘static’ (i.e. the room isn’t changing shape dynamically), and are linear and ‘consistent’ i.e. for every fluctuation in frequency response there is a corresponding delayed reflection which our ears pick up as delays in the time domain, not just the frequency magnitude. The whole thing hangs together as we turn our heads or move around the room. And of course the direct sounds from the performer to our ears are not affected by the room. I would contrast this with splitting the audio up into several frequency bands and applying arbitrary delays and phase shifts to each band. Does anything like this ever happen in nature? Is there any reason to think that our ears and brains have evolved to ignore, or cope with this? Does it, in fact, resemble an acoustic that could never exist, or a room whose shape is changing dynamically? Maybe we can’t consciously discern it, but is it a fatiguing effect nevertheless?
I know that there has been work done on the audibility of phase in the past, but I would ask the following:
(a) Was the audibility of ‘phase shift’ vs. ‘no phase shift’ tested? From what I have seen, it was usually ‘phase shift’ vs. ‘different phase shift’.
(b) What sort of music and recordings was it tried with? I think that this choice probably limits the usefulness of all listening tests (see my post on Science and Audio for my scepticism regarding listening tests!).
Here’s a link to an extremely interesting speaker review where the author tells us how messed-up phase unambiguously “colours” the sound of transients, even though the steady state frequency response can be perfectly flat: http://www.iar-80.com/page96.html
Another one of my pet subjects is bass reflex speakers. Can we state that we know what effect the room has on the bass if the speaker itself is doing strange, unnatural things with the bass? Are audiophiles spending inordinate amounts of time and money trying to fix the room with treatments and/or EQ when the “problems” are being compounded by an additional (unnecessary?) resonator?
I realise that my speakers are not a point source, but I did try to get the drivers as close together as possible. As I say in the article above, there are people who argue that speakers should be wide flat-fronted boxes, and others who say they should be omnidirectional. Mine are sort of in between, so they might be the perfect compromise, or the worst of all worlds! They seem pretty good to me, anyway.
Turning the question of flat phase response around: given that I can program in any phase response I want, what should that response be? I would think it obvious that I want the acoustic phase of the drivers to overlap perfectly as they cross over. But beyond that, do I want the overall response of the speaker to be flat, or ‘minimum phase’ or what? To me, the only ‘natural’ response is the flat (zero) phase (plus an inevitable delay). Other questions might override that decision, such as latency. In my system the latency is huge: about a second, so totally unsuitable for video work unless the video can be delayed too. I notice that other FIR-based linear phase systems are similarly burdened. The latency can be reduced, with the accuracy of the filters gradually being compromised, but it’s not something I have tried – yet. If the video can be delayed then I don’t think it’s an issue (except for the system’s responsiveness to the controls being operated..?)
LikeLike
“I would contrast this with splitting the audio up into several frequency bands and applying arbitrary delays and phase shifts to each band. Does anything like this ever happen in nature? Is there any reason to think that our ears and brains have evolved to ignore, or cope with this? Does it, in fact, resemble an acoustic that could never exist, or a room whose shape is changing dynamically? Maybe we can’t consciously discern it, but is it a fatiguing effect nevertheless?”
Very fair points and questions in my opinion. I’m not sure two point sources generating the information for a ‘centred’ audio images also happens in nature. That 1kHz test I will show is extremely ‘fatiguing’ to listen to. And by definition attempting to create a natural sound field with a L and R channel located anywhere other than directly over your ears yields an unnatural result in general. Binaural recordings seem to be the only way to reproduce the accurate sounds fields that your brain finds no fault with – in my experience at least.
example: https://www.youtube.com/watch?v=IUDTlvagjJA
“there has been work done on the audibility of phase in the past, but I would ask the following:
(a) Was the audibility of ‘phase shift’ vs. ‘no phase shift’ tested? From what I have seen, it was usually ‘phase shift’ vs. ‘different phase shift’.
(b) What sort of music and recordings was it tried with? I think that this choice probably limits the usefulness of all listening tests (see my post on Science and Audio for my scepticism regarding listening tests!).”
Very good questions. We both have the equipment to test and generate insight. Btw I notice that you blog has lots in the way of ‘thinking’ but it’s light on talking about your listening experiences and experimental investigations. The listening exploration would be very interesting. Especially when based on rational curiosities!
“so they might be the perfect compromise, or the worst of all worlds! They seem pretty good to me, anyway”
I’m not surprised at all that they sound fantastic. You hit all the main considerations and the remainder may not be that important. ‘Maybe’ side by side comparison would yield different results between designs – but what would be the point? You probably have more than what you need for a very pleasurable and visceral experience of an otherwise flawed two channel audio scene reproduction technology. That being said, the L-R technology can still be fantastic to listen to.
BTW I took my design don’t to *THE* hifi store in town and successfully put it up against some very expensive hardware. I rolled in with the amp in my pack back and a speaker in each hand. Hilarious. Mine didn’t sound conclusively *better* but for 500$ in parts there’s no way any rational person would spend even twice as much for the other stuff – unless they like ‘art’ pieces. Which they seem to.
“Turning the question of flat phase response around: given that I can program in any phase response I want, what should that response be? I would think it obvious that I want the acoustic phase of the drivers to overlap perfectly as they cross over. But beyond that, do I want the overall response of the speaker to be flat, or ‘minimum phase’ or what? To me, the only ‘natural’ response is the flat (zero) phase (plus an inevitable delay).”
Maybe it just doesn’t matter? What’s next, controlling room reflections (vertical flat services also being unnatural)? If anything, I think the next step is experimenting with the all the great questions above. Try stuff and see. I don’t think you can think your way out of it from here.
“Other questions might override that decision, such as latency. In my system the latency is huge: about a second, so totally unsuitable for video work unless the video can be delayed too. I notice that other FIR-based linear phase systems are similarly burdened.”
Well, now we can see the inherent FIR problem. My classic IIR DSP filters are just a handful of coefficients and very fast to calculate. I use my system for video. And generally I (and other users) would find a 1s latency to be intolerable when operating controls. It’s bad enough I have the ugly speakers in the living room. lolol. I don’t need any other inconveniences that may jeopardize the somewhat invasive project in my household.
However, I am still very interested in answers to your questions.
LikeLike
Maybe you can grab an OSX HDMI-enabled device and a cheap 7.1 HDMI AVR and try my IIR approach for comparison? Maybe you can run a Hackintosh? You may even appreciate the lack of wires and elegance in general. Not to mention that you get the full OSX video experience for free too. Just plug in a monitor. Or a projector.
Also, FWIW I have listened to many fantastic sounding hifi systems in recent years. Some DSP based. Some based on first order passive crossover filters. None of this crossover stuff is obvious except for the purely rational stuff we can think through. Like how I can generate 1200WPC (peak) of clean audio amplification from a raw digital signal using a 200$ commodity amplifier. Awesome. Can’t do that with a three way passive approach.
LikeLike
This is what I have implemented now in my system. I am using the very excellent and cheap (purchased mine for $55 off of eBay) Pioneer VSX-D912 receiver. My system uses IIR filters (by way of a MiniDSP 4x10HD). Tri-amped JBL speakers. 15″ 2226, 2″ compression driver (2445) and super tweeter (2405). Most of my listening is done at -40 dB on the receiver (lots of headroom to spare!).
I do like the idea of experimenting with the Mac-based system, though. Seems like a lot more flexibility to try and compare IIR/FIR-based crossover filters, linear/minimum phase etc.
@Andy, what Sony receiver do you use? My Pioneer does not have HDMI in it (hence the low price) but it is 100 watts/channel and has plenty of power to spare (especially with the high-efficiency horns).
LikeLiked by 1 person
” I’m not sure two point sources generating the information for a ‘centred’ audio images also happens in nature. ”
No, I suppose not, but it is wrong in quite a ‘simple’ way! If there are supposedly-horrendous things happening with phase shifts etc. as you move around the room then they’re not obvious as you listen. And the fact that both ears are receiving both speakers’ output doesn’t seem to stop you from getting a pretty good illusion of a “soundstage”. Each speaker may be individually correct, so in itself the result is two ‘correct’ speakers playing in unison within your fixed, static and consistent room acoustics. As I say, without proof of any sort, I imagine that this will be easier to listen to than speakers that are applying arbitrary frequency-dependent phase shifts. But I do realise that very few people in the audiophile world give this argument any credence at all. It makes me happy, anyway.
I recently listened to some ambiophonic recording examples (you have to push your speakers closer together) and they were pretty good. A deliberate attempt to make speaker-based stereo more ‘correct’ and, I suppose, ‘natural’. However, the algorithm to do this involves each speaker recursively producing attenuated, delayed, inverted versions of the other’s output signal. Refinements include limiting this to the mid-range. As I say it sounds good, but on the examples I heard it came as a relief to eventually go back to plain stereo. Maybe if I had a working system to do the correction myself I could find better settings.
Did you see that speaker review I linked to? It’s an amazing piece of work, and talks about something that I have rarely seen mentioned: the idea that unless all frequency components are lined up with the correct timing and phase, it will “colour” the sound of transients. By going with linear phase filters I can cross that potential problem off the list without doing any experiments.
“I notice that you blog has lots in the way of ‘thinking’ but it’s light on talking about your listening experiences and experimental investigations…”
Yes, that’s the “rational” aspect. I think we could spend the rest of our lives experimenting in semi-random fashion but not come to any conclusions. I am trying to do it the other way round: find the most blameless solution I can think of that is sufficiently practical for my purposes, implement it, and cross it off the list. It frees me up to then concentrate on the interesting stuff such as maybe putting together some omnidirectional speakers like yours…
LikeLike
“No, I suppose not, but it is wrong in quite a ‘simple’ way! If there are supposedly-horrendous things happening with phase shifts etc. as you move around the room then they’re not obvious as you listen.”
There are horrendous things happening. And your brain seems to deal with when listening to full spectrum sound with a L-R channel stereo system. It’s really quite amazing.
The video below shows full 0-360 deg phase interference from a supposed stereo image located between the two transducers. And this is with room reflections. It would be much more pronounced without the room influence. Higher frequencies result in closer spacing.
http://www.phys.uconn.edu/~gibson/Notes/Section5_2/Sec5_2.htm
http://video.mit.edu/watch/sound-wave-interference-8396/
Anyway, all that above makes me think that the crossover phase question may not be that relevant in the scheme of things. But I have no real conclusive arguments to support that.
“And the fact that both ears are receiving both speakers’ output doesn’t seem to stop you from getting a pretty good illusion of a “soundstage”.”
Yup, this is true. Lucky for us. We seem to filter *a lot*. Next time I’m told that I have selective hearing……
“Each speaker may be individually correct, so in itself the result is two ‘correct’ speakers playing in unison within your fixed, static and consistent room acoustics. ”
It’s still my observation that two ‘correct’ speakers in a room trying to create a centered image is an inherently flawed approach regardless of how static and consistent the environment is.
“I recently listened to some ambiophonic recording examples (you have to push your speakers closer together) and they were pretty good. A deliberate attempt to make speaker-based stereo more ‘correct’ and, I suppose, ‘natural’. However, the algorithm to do this involves each speaker recursively producing attenuated, delayed, inverted versions of the other’s output signal. ”
I know this research. I don’t really see the point for Hi-Fi. That binaural link I posted really takes care of everything in the best possible way – except for the sound you feel with your body and not your eardrums. My impression is that the researcher would like to get it into 3DTV applications where the lack of fidelity would be less of an issue. Or perhaps for iPad or laptop speakers. All huge revenue opportunities. hmmmm maybe we picked the wrong projects!
On the other hand I’ve experimented with introducing ‘cross talk’ for headphone listening. The process where you ‘place’ the speakers in front of you by strategically cross-feeding the signal. That’s really great an eliminates the centred image problems discussed above.
“Did you see that speaker review I linked to? It’s an amazing piece of work, and talks about something that I have rarely seen mentioned: the idea that unless all frequency components are lined up with the correct timing and phase, it will “colour” the sound of transients. By going with linear phase filters I can cross that potential problem off the list without doing any experiments.”
This is generally a slippery slope of acceptance. In the same way I could blindly use gold plated power cords so I don’t have to worry about how the copper oxidization influences the resultant sound. I suggest that generally there is more of a price to pay for FIR filters than IIR so it’s worth some investigation of their pros and cons. That being said, I don’t think we’ll find anything conclusive to make one approach more compelling that the other in any kind of audible way – especially given the above. But you never know.
I’m not sure about that article. It seems written from an emotional place and has all the hallmarks of audiophile kool-aid. I get what’s being said but there must be a better way to evaluate. Besides the listener really has no way to know what he is listening to. He thinks it might be phase colouration. Perhaps it’s something completely different. Perhaps it’s nothing? No real way to know. I drink plenty of kool-aid myself from time to time tho. It tastes good going down.
“Yes, that’s the “rational” aspect. I think we could spend the rest of our lives experimenting in semi-random fashion but not come to any conclusions.”
Who said anything about semi-random? And experiments don’t necessarily have to be time consuming. It’s just more about looking for answers in the physical world rather than on the internet.
“Find the most blameless solution I can think of that is sufficiently practical for my purposes, implement it, and cross it off the list.”
This is really the best approach. Are the FIR filters for you any more practical than the more efficient IIR filters?
On an unrelated characteristic of the Omni design: the two way version can have outstanding bass if you don’t want to listen too loud. The travel of the mid-woofer is fantastic and the addition of some digital bass lift really produces dramatic results. Again, this would be highly impractical with a passive crossover.
I have about 10 more of those bass tubes sitting around since I was forced to buy an entire 20′ green sewer pipe. I can send you (anyone) some if you (they) pay for shipping 🙂
A
LikeLike
“This is generally a slippery slope of acceptance. In the same way I could blindly use gold plated power cords so I don’t have to worry about how the copper oxidization influences the resultant sound. I suggest that generally there is more of a price to pay for FIR filters than IIR so it’s worth some investigation of their pros and cons. That being said, I don’t think we’ll find anything conclusive to make one approach more compelling that the other in any kind of audible way – especially given the above. But you never know.”
The difference being that if we tried to measure the effect of the gold plated power cord we would not detect anything at all because its effects would be less than 1 LSB of a 20 bit signal – probably, if I had to bet on it. But the phase thing is unambiguously measurable. The question is whether is it audible or not. Maybe it is not audible 99% of the time. Or not audible at all on certain types of music.
There is one listening test where I can always identify the difference between ‘correct’ and ‘otherwise’ when switching between the two. If I play noise through the system and turn off the correction I can hear the effect. This isn’t just a test of overall phase correctness, however. This is the difference between a suckout and no suckout because the correction is affecting the relative phase between the drivers as well as the overall phase linearity. Applying the correction, it is clear that a notch in the response is being ‘filled out’. Similarly, adjusting the relative delay between the drivers it is clear when they are best aligned at the listening position and therefore cancelling each other out the least. With certain types of music this is far from clearly audible. It seems to be most obvious on music which features a ‘wash’ of overdriven electric guitar. But my anecdotal subjective experience is that after a while without the correction, there does seem to be an edginess or hardness to the sound regardless of the type of music. Such a statement is not “scientific”, though, which is why I simply use the corrected FIR filters and forget about it. The latency of the large FIR filters I have chosen to use doesn’t bother me. Two years in, and happily living with a second’s latency – it is a lot less than the latency of vinyl!
LikeLike
“The difference being that if we tried to measure the effect of the gold plated power cord we would not detect anything at all because its effects would be less than 1 LSB of a 20 bit signal – probably, if I had to bet on it. But the phase thing is unambiguously measurable. The question is whether is it audible or not. Maybe it is not audible 99% of the time. Or not audible at all on certain types of music.”
Fair argument. However, I can measure the increase in oxidation of the plug contacts to a much more significant effect than that. Also, I can measure the difference between the inductive and capacitive nature various cords as well. Is it audible? I know what I think about that…
“There is one listening test where I can always identify the difference between ‘correct’ and ‘otherwise’ when switching between the two. If I play noise through the system and turn off the correction I can hear the effect.”
If you play noise and switch between *anything* you will hear a difference. I do this with my time corrections and various other filters etc. A difference between one or the other is very clear with noise – much more clear than with music. It’s probably subjective as to if it’s filling a ‘notch’ or not; that’s a qualitative assessment. Playing music and assessing which is more accurate or images better or sound less smudged is very difficult for me; even if I take settings way out of where they *should* be to make rational sense. The mind is a funny thing.
This afternoon I started a new room assessment technique. A 10Hz audio square wave generator and a listening microphone fed into a scope. So far so good. Forget the freq-domain stuff 🙂 I’ve never looked at my system like this before in the time domain. I don’t know why.
“while without the correction, there does seem to be an edginess or hardness to the sound regardless of the type of music.”
Interesting. Wonder if it’d be the same with IIR filters?
“it is a lot less than the latency of vinyl!”
Because you have to change records?
LikeLike
“I can measure the increase in oxidation of the plug contacts to a much more significant effect than that. Also, I can measure the difference between the inductive and capacitive nature various cords as well.”
Of course we can measure these, but do they actually have any measurable effect on the audio? I have noticed that cable enthusiasts (not implying that you are) often refer to the widely-varying parameters of L, C and R between different cables, but don’t take it to its conclusion and show how these would affect things in the context of the circuit. They fail to show that increasing the resistance of a signal cable by a factor of a thousand might reduce the volume by 0.0001 dB, or whatever, and that its effects are predictable and benign. In terms of oxidation, if the power cord contacts are ‘wiped’ every time the plug is inserted, does the oxidation actually have any effect at all?
“It’s probably subjective as to if it’s filling a ‘notch’ or not; that’s a qualitative assessment. ”
I am not saying that I can listen to a single sample of noise and say whether the response is flat or not or in what way it is not flat, but I can clearly hear the difference when the notch (that in this case I know is there) is filled in – and comparing the two conditions side by side I could tell you which was which.
“Interesting. Wonder if it’d be the same with IIR filters?”
Well the implication is that if you measure your drivers’ raw phase responses and they are mismatched through the crossover, then even if you use crossover filters with perfectly-matched phase you will still get an acoustic suck-out. The benefit of measuring the system from end to end is that you automatically compensate for all the sources of phase shift and, in conjunction with the adjustable delay, you can set the system to something close to optimum. You can always say “Ah, but is it optimum? Can you prove that suck-outs aren’t beneficial to the sound?” etc. but in this case I am happy to go for the straightforwardly-obvious setup.
“Because you have to change records? ”
Of course. And get the LP out of its sleeve, carefully place it on the platter, cue up the arm, wait while it drops onto the run-in, and wait for a few seconds for the music to start. The “latency” and lack of responsiveness of the controls is huge!
LikeLike
@danvprod
My AVR is very much like this one but I think the previous model:
http://wwv.crutchfield.com/p_158STDH520/Sony-STR-DH520.html
LikeLike
I’m not a cable enthusiast. I was just making the point that one can ‘cast doubt’ of throw irrelevant monkeys in to any technical argument. Very good points wrt the FIR approach. I may have to revisit this. Thanks.
LikeLike
What do we think about something like the LX Mini by SL? Seems to share the same mid-bass driver as @Andy’s omni loudspeaker system. The woofer is in a sealed enclosure, it is bi-amped, the system uses DSPs and SL recommends a sensible 4-channel solid state amp. It’s a 2-way, not a 3-way but other than that, it seems pretty sensible to me (and priced quite nicely!). I’ve been on the fence about building that setup, or a system more similar to what you have (3-way sealed boxes with a large LF driver).
LikeLike
My design is a modified SL Pluto – the predecessor of the mini. I’ll modify mine to the LX Mini tweeter in due time. As I mentioned the LX Mini bass is surprisingly impressive as a bass lifted two way. It’s probably fine for most rooms unless you like it quite loud. I added the 10″ woofers because a) I like it loud, b) i like experimenting (I can compare the two), c) I had 7 amp channels anyway, d) I needed a wider base so it would stand up properly.
LikeLike