I beg to differ. I run detailed (read: high resolution) filters that are completely linear phase (wouldnt be able to get the flat phase response I have without them)... as I previously stated I run many filters with these qualities on a rather old system at less than 50% utilisation of resources. A modern system based on intels core architecture has probably 3-4 times the raw crunching ability of the dated platform my system is composed of and you say that this "plugin" consumes nearly 100% of resources?
I dont understand. So you have a monumentally overpriced "prosumer" external sound card (with no doubt, the same well speced processing/dac chips that most soundcards use) and a squeezebox? Where does the squeezebox come in? Dont tell me you use digital outs from the RME device to send data to the squeezebox?! How is the squeezebox a "pure digital device" exactly? Its an embedded system that can decode audio formats and send them to an onboard dac... unless your just using it as a buffer in your audio chain, then its near pointless....
This has now gone a little too far off topic I guess.... I'd love to continue the discussion of certain peoples idiocy with regard to computer based hi-fi over in the computer section... god knows it could use some real debate, the SNR over there is ridiculous.
I've some suggestions for the thread in the PC based forum. As it doesn't seem to have been started yet I'll post them here. Successful technology evangelists show people pathways for using the technology instead of complaining about other folks choosing to do things differently.
Among those numerous freeware programs are there any you'd suggest as easy to use for crossover and EQ design? It would be great to see a thread illustrating an accessible end to end workflow for how to get this stuff running. Preferably on a PC or Mac, or at least cygwin, and preferably one that has a reasonable chance of working out of the box---in particular, I'm curious how you'll address device driver availability for audio interfaces.
Recognize that folks who aren't electrical engineers have a substantial learning curve to deal with something like Octave or Matlab (not to mention my experience with Octave being it crashes so often as to be unusable and Matlab only being of use if you have the filter synthesis toolbox and a version that will run on your OS). Also recognize most folks are more interested in getting results than in tinkering with an OS. Unix variants, Linux included, tend to cater more to the tinkering mentality and, unless you have a system administration background, are not particularly accessible to the average user. This is coming someone who spent years doing network and system administration on everything from NextSTEP to OSF/1 to Solaris to Linux to Windows 3.0 up through XP. You and I can hack it, but unless you can articulate a value proposition for brutefir on Linux that's cost effective in terms of both time and money you're unlikely to see any significant adoption.
That said, you can find my posts earlier in this thread where I point out PLParEQ's relatively low return on value.
This workflow seems to be missing some steps. I assume you're referring to DSPreLab and not HOLMImpulse as that's where the crossover and equalization functionality resides but so far as I know DSPreLab lacks the ability to export FIR coefficients to something other than the DSPre. Like, say, a coefficients file for BruteFIR or SigmaStudio.
Among those numerous freeware programs are there any you'd suggest as easy to use for crossover and EQ design? It would be great to see a thread illustrating an accessible end to end workflow for how to get this stuff running. Preferably on a PC or Mac, or at least cygwin, and preferably one that has a reasonable chance of working out of the box---in particular, I'm curious how you'll address device driver availability for audio interfaces.
Recognize that folks who aren't electrical engineers have a substantial learning curve to deal with something like Octave or Matlab (not to mention my experience with Octave being it crashes so often as to be unusable and Matlab only being of use if you have the filter synthesis toolbox and a version that will run on your OS). Also recognize most folks are more interested in getting results than in tinkering with an OS. Unix variants, Linux included, tend to cater more to the tinkering mentality and, unless you have a system administration background, are not particularly accessible to the average user. This is coming someone who spent years doing network and system administration on everything from NextSTEP to OSF/1 to Solaris to Linux to Windows 3.0 up through XP. You and I can hack it, but unless you can articulate a value proposition for brutefir on Linux that's cost effective in terms of both time and money you're unlikely to see any significant adoption.
That said, you can find my posts earlier in this thread where I point out PLParEQ's relatively low return on value.
Some of the reaction you've gotten here is due to presenting as an OS fanboi. Taking the time to understand ASIO would help in that regard.
Theo:
The squeezebox is my digital playback source. The digital output of the Squeezebox is fed into the Fireface, which sends the datastream through the filters that run in the computer, and then back to the Fireface on eight channels (4-way XO). The reason I choose the Fireface is that is has a unique combination of features found in very few (if any) devices on the market - like mic inputs, instrument line input, and great internal routing matrix and mixer. The only thing I'm missing is AES3 interface. I didnt buy the Fireface for its DAC's ---- they are not high-quality, but as a computer interface its great I think. To make it even worse, I have an external DAC in addition to the Fireface....
What I love with manually dialed-in classic EQ instead of generated FIR filters is that I can adjust to taste in real-time, sitting in my listening chair. Perfectly flat measured response is something I usually dont like too much either. To create a 4-way system with the PLparEQ plugin I need to run at least four instances of the plugin, that will generate four times the CPU load. Thats an other explanation for my high CPU load.....
Yes this *is* rather OT in this thread....
The squeezebox is my digital playback source. The digital output of the Squeezebox is fed into the Fireface, which sends the datastream through the filters that run in the computer, and then back to the Fireface on eight channels (4-way XO). The reason I choose the Fireface is that is has a unique combination of features found in very few (if any) devices on the market - like mic inputs, instrument line input, and great internal routing matrix and mixer. The only thing I'm missing is AES3 interface. I didnt buy the Fireface for its DAC's ---- they are not high-quality, but as a computer interface its great I think. To make it even worse, I have an external DAC in addition to the Fireface....
What I love with manually dialed-in classic EQ instead of generated FIR filters is that I can adjust to taste in real-time, sitting in my listening chair. Perfectly flat measured response is something I usually dont like too much either. To create a 4-way system with the PLparEQ plugin I need to run at least four instances of the plugin, that will generate four times the CPU load. Thats an other explanation for my high CPU load.....
Yes this *is* rather OT in this thread....
How long ago did you get the Fireface? This feature set is now normal in midrange interfaces (USD ~300). Granted, you'd have to spend more than that to match the box spec of RME's FPGA, but that's more than is required for running a four way crossover.
It's all about latency. FIR is clearly very expensive compared to IIR, but BruteFIR makes it manageable by trading off cpu utilization for latency. The Windows tools widely discussed here are for music creation not for general signal processing, and so generally need very short latencies. Crank your BruteFIR buffers down to 64 or even 32 samples and see what happens to your Athlon.
Having said that, for a music-only system there is no reason that a delay of 250ms or so is a problem, it's just that this isn't what most VST plugins are designed to do. I believe that convolver-vst manages this well, and it's free to boot.
I'm a long-time Linux/dsp xover tinkerer/experimenter, and even after 10+ years playing around with this stuff I think this is still unfortunately the crux of the problem. Theo isn't really incorrect in saying that some aspects of this problem are 'easy' in Octave/Matlab, but I think it's pretty telling that despite how 'easy' some of this is we still don't have anything like a usable freely-available filter generation package that the 'average joe' can use. I've started looking at tools again after a couple years on the sidelines, and for FIR filters it really appears that it's (((accourate))) or 'nothing' (i.e. hand-coded Octave scripts). I'm to the point that the ~$400 or so that accourate would cost me is starting to look like a pretty decent bargain despite the fact that I am "fully capable" of writing the Octave code to do what I want - it's patently clear that I will never take the time to do so, so the fact that I 'could' do it is a moot point.
I *did* find what looks to be a good Biquad designer in the form of the Hypex filter designer. This is aimed at exporting the results to their DSP amp products, but it is basically a graphical Biquad editor that allows importing of driver impulse responses and shows the individual filtered responses and the summed responses. It does not support Target responses, but it's better than anything else I've seen so far.
Of course, you still have the problem of translating these filters (which are Q based) into the flavor of biquad that your EQ uses - the Reaper eq for example uses 'Bandwidth' rather than Q, and it doesn't even look (at first glance) that B = 1/Q.
Fortunately, my perspective on EQ has changed as well. I used to be a Linear Phase guy, but my opinion now is that IIR is the appropriate topology for all crossovers and basic driver eq, and that only a single global FIR correction stage is really needed. I see the availability of things like the Dueland topology and the phase-coherent subtractive iir topologies more interesting at this point than a brute-force linear phase FIR crossover. Remember that Danley gets phase-coherent response out of his Synergy designs using purely passive xovers.
Unfortunately, this is a perspective that I've grown to mostly agree with. I'm a 'Unix guy', having done fairly low-ish level Unix software dev since college and have always been a big Linux user. Even for me though, I've finally moved away from Linux for audio work simply because the overhead of managing the system and the dearth of good tools is just too much if all you care about is the results. The final straw was struggling to get my 'supported' firewire interface (Saffire Pro 24) working - it 'almost' works, but not quite. Conversely, in about 30 minutes of hands-on time or so I was up and running with Reaper, ReaRoute, Foobar and HolmImpluse with the ability to dial-in an IIR xover.
Now, the MPD + BruteFIR combo IS the best PC runtime playback platform that I've seen by a good margin - IMHO it's either that or a Squeezebox - anything else is cumbersome. I 'intend' to return to it once everything is settled, but the 'development workflow' is tedious when you're primarily experimenting with things.
4-5 years ago I think... long time.
I have yet to find a cheaper interface that works with laptops, that has either ADAT or AES3 for multichannel digital I/O as well as multichannel analog I/O, SP/DIF I/O, mic input and the routing software of the RME. I think its *very* flexible.
Indeed - if 4-way XO with analog out was the only requirement, there are lots of other devices to choose from.
This is an interesting conversation. Could you explain a little more of this in lay terms? I've just started playing with linear phase software, so I'd like to know more, but I just don't know much about all of these different ways of doing things. Are you saying you'd use a single FIR filter to get things phase linear, and let IIR handle the rest of the XO/EQ? I'm just guessing here...
Assuming you have something useful to share, nobody is going to listen to you with an attitude like this. Its immature, and it is exactly this sort of approach that lower the SNR.
Last edited:
I hit the same point with Linux for my personal stuff back in 1998, though I continued to use various unixen for work up through 2004. Good data, thanks, especially as I'm using a Saffire as well. If I didn't have a requirement for linear phase I'd just use a Motu Ultralite but, as it stands today, the ~$125 incremental cost riser for a linear phase PC crossover's quite attractive compared to fixing bugs in third party Linux device drivers.
From my investigation Duelund's a major hassle to implement with dipoles due to the limited agility in crossover points and slopes. Implementing a four way linear phase crossover on the PC is much easier.
Is Danley using Dunlavy's approach or something else?
Cumbersome how? I've found Bidule+Allocator+PLParEQX3 quite easy to work with.
That's certainly one option. It'd be interesting to see how CPU load and latency compared to efficient time reversed IIR implementations like Allocator or FFT based approaches like PLParEQX. I'd expect IIR+FIR to win on latency and lose on CPU and usability (not that the IIR part is particularly difficult---Allocator Lite's all of USD 60).
You just described the Saffire Pro 40 I eBayed for USD 350 (new, shipping included) a couple months ago. In comparison the Fireface 800 is around USD 1500 on this side of the pond (the 400 is about USD 1000).
I see a lot of designers succumb to this idea of using a minimal baffle (or no baffle at all.) The problem isn't the baffle; the problem is the transition from the radiator to the room.
In other words, you can use a large baffle if you create a smooth transition from the radiator to the room that it's radiating into.
The sharp edge of an unterminated loudspeaker creates a lot of diffraction, and diffraction effects are very audible.
In general I agree - but (I'm guessing) that smoothing diffraction may be a design aspect of boxed speakers. OB seems like the opposite - it has 'maximum' diffraction - and the quality of the final acoustic response depends on maintaining as regular a front and rear source so both sides diffract the same and one ends up with a regular polar response.
Something I've wondered about box speakers is about the baffle step region - if the box is big enough, the step can be pushed into the low hundreds of Hertz, and perhaps that is sonically a benign place to begin beaming (which is what Earl has done, and is also seen in the Poor Man's Strad). Basically, what I'm saying is bigger is better when it comes to boxes, at least in terms of directivity. But thats not my area at all.
Something I've wondered about box speakers is about the baffle step region - if the box is big enough, the step can be pushed into the low hundreds of Hertz, and perhaps that is sonically a benign place to begin beaming (which is what Earl has done, and is also seen in the Poor Man's Strad). Basically, what I'm saying is bigger is better when it comes to boxes, at least in terms of directivity. But thats not my area at all.
So your 2x4 with a naked driver strapped to it achieves this?
You think large diffraction sources magically cancel each other out?
Just in case you didn't read it yet, Siegfried Linkwitz has recently published some interesting measurements of and reasoning about diffraction in OB speakers: Diffraction from baffle edges
It mainly shows that baffle edge diffraction is highly frequency dependent and that baffle symmetry is clearly detrimental, so nothing really new there; nonetheless I found it very illuminating to see those points so clearly demonstrated.
Which brings me to the work you've done, cuibono: I'm greatly impressed with your relentless research and testing of concepts! I' ve already learnt a lot from following the discussion and studying your experiments and measurements, and I think with the results you've obtained so far, you have really pushed the limits for (not only) DIY speaker concepts. A huge thanks for letting us share so much about your efforts!
And, having never been much of a DIY guy, I might even start becoming one thanks to the Violet...
Wolfgang
It mainly shows that baffle edge diffraction is highly frequency dependent and that baffle symmetry is clearly detrimental, so nothing really new there; nonetheless I found it very illuminating to see those points so clearly demonstrated.
Which brings me to the work you've done, cuibono: I'm greatly impressed with your relentless research and testing of concepts! I' ve already learnt a lot from following the discussion and studying your experiments and measurements, and I think with the results you've obtained so far, you have really pushed the limits for (not only) DIY speaker concepts. A huge thanks for letting us share so much about your efforts!
And, having never been much of a DIY guy, I might even start becoming one thanks to the Violet...
Wolfgang
Our eye is working in a bandwith of one octave: 0.35-0.7 µm roughly.
Our ear is working in a bandwith of nine octaves - at least.
That's why our eye has no idea about 'sharpness' being a function of wavelength. But you can not talk about a 'sharp' edge in acoustics without relating it to the wavelength of the frequencies you are talking about.
That said, edge diffraction in acoustic dipoles doesn't hurt as long as the distance from the acoustic source to the diffracting edge is kept below half the wavelength of the frequency in mind. You simply loose SPL, but there is no phase change or time delay that your ear could detect.
If you keep to above rule, it does not matter how the 'diffracting' edge is shaped. Only if you decide to choose a shorter wavelength or wider baffle, diffraction might get you into trouble. But by smoothing those troublesome edges you don't avoid diffraction, you just smear it over a wider frequency and time band.
That's how I look at it.
Rudolf
Our ear is working in a bandwith of nine octaves - at least.
That's why our eye has no idea about 'sharpness' being a function of wavelength. But you can not talk about a 'sharp' edge in acoustics without relating it to the wavelength of the frequencies you are talking about.
That said, edge diffraction in acoustic dipoles doesn't hurt as long as the distance from the acoustic source to the diffracting edge is kept below half the wavelength of the frequency in mind. You simply loose SPL, but there is no phase change or time delay that your ear could detect.
If you keep to above rule, it does not matter how the 'diffracting' edge is shaped. Only if you decide to choose a shorter wavelength or wider baffle, diffraction might get you into trouble. But by smoothing those troublesome edges you don't avoid diffraction, you just smear it over a wider frequency and time band.
That's how I look at it.
Rudolf
Wolfgang - you're welcome! I really enjoy sharing, and having a community to do so, and discuss with others is one of the big things that pushes me. I've learned almost everything I know here at diyAudio. Perhaps in the future, I'll try and spell out more about how to construct the Violet's, to make it easy for others.
I agree with Rudolf here - and like I said above, it seems diffraction is not as much of an issue for OB speakers as it is for box speakers. I wonder what John K has cooking - I think he might surprise us with something related before too long.
I agree with Rudolf here - and like I said above, it seems diffraction is not as much of an issue for OB speakers as it is for box speakers. I wonder what John K has cooking - I think he might surprise us with something related before too long.
This is one of those things that's always baffled me. (get it? haha)
Here's the deal -
It's difficult to see the effects of diffraction unless you do polar measurements. Most DIYers don't do polar measurements. Therefore, most DIYers aren't aware that a few dollars worth of PVC pipe can clean up the power response of their loudspeakers.
While I applaud the work of Linkwitz, I believe that he dismisses the effects of diffraction too quickly. And it would be great if he started publishing polars. IMHO, the power response of a loudspeaker is more important than the on-axis response. And again, a few bucks worth of PVC cleans up the power response in a huge way. It's not an expensive or difficult enhancement.
Last edited:
I've taken off-axis measurements of my bare Neo3PDR. Others have posted polar measurements on this forum as well. It's difficult (for me) to interpret my results. With no edge treatments, there's a sharp null around 8kHz, which corresponds to 1/2 the Neo3's width, so I think that's expected. But the off-axis curves track the on-axis response up to around 6kHz. ARTA's calculated averaged power response is fairly flat.
Putting 1" PVC pipe or wool felt on the edges smoothes out that null a lot. However, that seems to impact the off-axis response, which now starts to deviate from the on-axis response around 4kHz. And the calculated power average gets worse.
So which is more important, smooth on-axis response or regular off-axis response? For now I'm listening with no edge treatments, speakers pointed forward so my listening axis is around 20-30 degrees to the side, which is where the on-axis null starts to even out.
Details here if you're interested, starting at around post 199:
HTGuide Forum - Getting back to my OB design...
Putting 1" PVC pipe or wool felt on the edges smoothes out that null a lot. However, that seems to impact the off-axis response, which now starts to deviate from the on-axis response around 4kHz. And the calculated power average gets worse.
So which is more important, smooth on-axis response or regular off-axis response? For now I'm listening with no edge treatments, speakers pointed forward so my listening axis is around 20-30 degrees to the side, which is where the on-axis null starts to even out.
Details here if you're interested, starting at around post 199:
HTGuide Forum - Getting back to my OB design...
Last edited:
Your measurements illustrate something that I've found in my own experiments. At some frequencies, the diffraction treatment is improving things. At others, it's no better, or even worse.
This gets into one of the tricky aspects of diffraction control, which is that you have to factor in the shape of the cone, and the transition from the radiator to the edges.
One things that's been a tremendous help for me is clay. I get these big rolls of clay from Home Depot, and apply them so that that the transition from the cone to the baffle is uninterrupted.
"The devil is in the details" as they say. I think you're on the right track; your measurements clearly illustrate that edge treatment makes a difference. At the same time, they illustrate that it's not as simple as slapping some PVC on the edge of the baffle. You have to maintain a smooth transition from radiator to edge. I'll be sure to acknowledge that in my future rants about diffraction
Saurav - you're seeing exactly why I don't apply edge roundovers. Its been said before, but this is the perfect example - by applying a roundover, you make the 'baffle' wider, which brings down the dipole peak in frequency, which then causes things start to go funny at 4kHz (versus 6-8kHz). Yes, roundover helps smooth the response some, but you also loose bandwidth, in terms of regular polar response.
For me, getting the most usable, regular bandwidth per driver is a high priority. Similarly, I prefer a notch at 8kHz rather than irregular polar response above 4kHz - I've found that the higher frequency errors are less noticeable. YMMV.
I like clay too though! In the past, I've applied it to various drivers, including the Neo3pdr - but it just lowers the dipole peak in this application. What I do with my Neo3's is a strip of foam at the edge - it doesn't bring down the dipole peak, but it helps smooth the 8kHz dip. There are pictures andmeasurements here somewhere - look in the thread I started about midranges. There is a link to it at the beginning of this thread.
For me, getting the most usable, regular bandwidth per driver is a high priority. Similarly, I prefer a notch at 8kHz rather than irregular polar response above 4kHz - I've found that the higher frequency errors are less noticeable. YMMV.
I like clay too though! In the past, I've applied it to various drivers, including the Neo3pdr - but it just lowers the dipole peak in this application. What I do with my Neo3's is a strip of foam at the edge - it doesn't bring down the dipole peak, but it helps smooth the 8kHz dip. There are pictures andmeasurements here somewhere - look in the thread I started about midranges. There is a link to it at the beginning of this thread.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.