Eh? I'm looking at instruction set and see no mention of a 56 bit word. 54 bit instruction word, 48 bit data word for the DSP. 76 bit accumulator, yes.
I agree the TAS3108 is a nice part and also fail to understand why TI is discontinuing it. Guess not enough folks understand the value of having more than 32 bits.
Thanks for mentioning this. I'd wondered about using delay to offload phase shift computations but hadn't gotten around looking into the synthesis.
Except concentric (coaxial) drivers like Tannoy, TAD, Cabasse, Beyma or Altec Lansing. Not to forget the APPOLITO 's configuration.
But, at this time no digital delays where available to recover the physical delay due to the non horizontal alignment of the coils. Nowadays, we can correct this in a nice way.
We have to notice this delay problem is important at the crossover. This argue (once more) in favor to 2way horns systems, where the crossover is at low frequency (<1000Hz) , so, phase correlation will not change a lot with vertical misalignment, because the wave length is 0.34m.
Last edited:
Of coarse coaxial drivers have lesser problems, I owned several Tannoy's. But most of the time they also need time alignment because their high drivers are not in the same acoustical centerpoint of the woofer due their construction.
But you can also overcome this vertical misaligment by using steeper slopes, the higher the order of the filter the smaller the frequency band where problems occur. I think using 48db filters has more effect than chasing the Fir filter holy grail.
Hi ds23man,
What would be the benefit of using asymmetrical slopes for time alignment of drivers vs. a delay on the tweeter? I can see the downside (bad phase tracking) but not the upside..
Taking inspiration from the Grimm Audio LS1 I would (am) use(ing) FIR only for phase EQ and IIR filters for the crossovers. Resulting in a linear phase sytem. (rePhase and a miniDSP openDRC are my weapons of choice for phase EQ at the moment, combined with Hypex PSC2.400d based IIR XO).
Just trying to understand why you make such a choice.
Edit: (Big) typo
What would be the benefit of using asymmetrical slopes for time alignment of drivers vs. a delay on the tweeter? I can see the downside (bad phase tracking) but not the upside..
Taking inspiration from the Grimm Audio LS1 I would (am) use(ing) FIR only for phase EQ and IIR filters for the crossovers. Resulting in a linear phase sytem. (rePhase and a miniDSP openDRC are my weapons of choice for phase EQ at the moment, combined with Hypex PSC2.400d based IIR XO).
Just trying to understand why you make such a choice.
Edit: (Big) typo
Last edited:
I do not use asymmetrical slopes, it is mentioned in the article. What I want to point out is the fact that even if you use a phase linear xover, the problems with vertical misallignment still exists and is clearly shown in the polar plot. A lot of people make the mistake that if you use a linear xover, you will get a phase linear system. Well you can actually, but only in one listening position and you will need to strap your head to a headrest!
I do not use asymmetrical slopes, it is mentioned in the article. What I want to point out is the fact that even if you use a phase linear xover, the problems with vertical misallignment still exists and is clearly shown in the polar plot. A lot of people make the mistake that if you use a linear xover, you will get a phase linear system. Well you can actually, but only in one listening position and you will need to strap your head to a headrest!
Yupz, you`re quite right, all the usual design parameters have to be taken in account, time alignment being 1 of them, dispersion / lobing is another big one. The phase EQ (making the system linear phase) is just the icing on the cake.
Most of the time, the loudspeaker adds its own slope to the one of the filter. Not in the same way on the basses and trebles units. So, searching for symmetrical acoustical slopes drive-you most of the time to asymmetrical electronic/electric slopes.
You want to use bass speaker in the range where it does not fractionate. By habit, you will see a peak of its response curve just where you will cut-it. So your filter will have to cancel-it. No rules: each driver need a dedicated filter for the same cut-off slope.
In principle this is true, but with the current DSP's it's not difficult to first EQ the driver flat well beyond the XO points and apply an XO after that. This solves the problem you descibe here
.
How can-you EQ a horn slope ? Better use-it, no ?In principle this is true, but with the current DSP's it's not difficult to first EQ the driver flat well beyond the XO points and apply an XO after that. This solves the problem you descibe here
In principle this is true, but with the current DSP's it's not difficult to first EQ the driver flat well beyond the XO points and apply an XO after that. This solves the problem you descibe here
And is the advised way of working by Hypex and other dsp manufactures.
Using asymetrical slopes is implementing the old way of working with passive xovers.......
As long as the system is minimum phase it should work, although you would need a lot of gain (which is probably knocked down again by the XO). But I am by no means an expert on horn systems. Should be an interesting thing to try!
If I were to implement an analog crossover (active or passive) I would use the driver's natural response as part of the transfer function.
If I would implement a customised digital crossover I would also include the driver's response into the transfer function.
If I were to use a turnkey solution like the Hypex crossover(s) then the easiest way is definitely EQing it flat and using a "texbook" crossover.
If you want to do this with reasonable accuracy then there is no way around measuring the horn's response. If there is no EQ function allowing an accurate compensation of the amplitude response then go for the one that gives the most accurate phase response. It is usually the phase response beyond the crossover frequency that influences the summed amplitude response the most.
Regards
Charles
If I would implement a customised digital crossover I would also include the driver's response into the transfer function.
If I were to use a turnkey solution like the Hypex crossover(s) then the easiest way is definitely EQing it flat and using a "texbook" crossover.
If you want to do this with reasonable accuracy then there is no way around measuring the horn's response. If there is no EQ function allowing an accurate compensation of the amplitude response then go for the one that gives the most accurate phase response. It is usually the phase response beyond the crossover frequency that influences the summed amplitude response the most.
Regards
Charles
Complete n00b here, wondering:
1. Has this progressed to the point where it could replace a passive crossover in an arbitrary speaker (say a B&W 801 Matrix Series 3) with an added (rented reference) microphone and computer (assuming no additional phase or room correction)?
2. As a pure DAC, is it comparable to existing consumer ones (say a Musical Fidelity M1a DAC)?
3. Assuming all digital input, is it possible/practical to compensate for phase problems offline, by statically pre-processing the files in advance? I played with Sox a long time ago, but I'm sure there are better options now.
I also used an NAD 118 a long time ago, and I always associate SHARC with lots of noise and distortion as a result. It did have a nice expander, though. Not sure if there's anything like it on this, or decent algorithm in the public domain.
I'm looking to build an NC400 and stumbled across this. The DLCP looks like it would make a nice integrated amplifier in combination, although its close enough to the edge to make me wonder if I'd end up with more than bloody fingers for trying.
Thanks,
Mark
1. Has this progressed to the point where it could replace a passive crossover in an arbitrary speaker (say a B&W 801 Matrix Series 3) with an added (rented reference) microphone and computer (assuming no additional phase or room correction)?
2. As a pure DAC, is it comparable to existing consumer ones (say a Musical Fidelity M1a DAC)?
3. Assuming all digital input, is it possible/practical to compensate for phase problems offline, by statically pre-processing the files in advance? I played with Sox a long time ago, but I'm sure there are better options now.
I also used an NAD 118 a long time ago, and I always associate SHARC with lots of noise and distortion as a result. It did have a nice expander, though. Not sure if there's anything like it on this, or decent algorithm in the public domain.
I'm looking to build an NC400 and stumbled across this. The DLCP looks like it would make a nice integrated amplifier in combination, although its close enough to the edge to make me wonder if I'd end up with more than bloody fingers for trying.
Thanks,
Mark
possible yes, practical... not so much as you need to process all your files, and find a way to keep your metadata (FLAC) in the process.
What media player software are you using?
Most of them have plugins that can do convolutions. I would not use a PC for a complete active filtering solution (for both practicality and reliability), but for doing only phase linearization and/or minor addtional EQ I think those plugins embedded in a mediaplayer do the job quite nicely.
Last edited:
Logitech SqueezeBox SB3 player, Win7 box with 8 cores and 32gb of memory. To really do it right, I think you'd need at least 80 bit floating point precision to keep 20 to 24 bit resolution all the way down the line, which is tough to do in real time, but easy to do offline: just process all the files and store them on the music server.
Going from Flac to a raw word stream, to a reversed stream, filtering and mixing the two, and back to Flac is pretty easy with sox. I don't have the scripts to do it any more, but it wasn't that hard. Even after recompiling it to use double precision (64 bit) though, I couldn't reverse the process and get back to my original. If I understand it right, that should be possible with all IIR filters and adequate precision.
Going from Flac to a raw word stream, to a reversed stream, filtering and mixing the two, and back to Flac is pretty easy with sox. I don't have the scripts to do it any more, but it wasn't that hard. Even after recompiling it to use double precision (64 bit) though, I couldn't reverse the process and get back to my original. If I understand it right, that should be possible with all IIR filters and adequate precision.
You don't need that much precision for a convolution (even with an FFT convolution).
foobar2000 convolver uses a 32bit float native format, whereas Jriver convolver uses 64bit, but I doubt it makes any difference.
I don't think SoX' offline convolution engine is any better (nor does it need to be).
foobar2000 convolver uses a 32bit float native format, whereas Jriver convolver uses 64bit, but I doubt it makes any difference.
I don't think SoX' offline convolution engine is any better (nor does it need to be).
The last SoX build I used was a while back but, if it hasn't been updated to support batch processing and metadata preservation, Cross Time DSP implements most of the necessary bits so far as FLAC goes. No takers yet on the last piece, though.
as there is no manual yet I was wondering if anyone can help - does the dlcp filter designer automatically center the response when all driver responses are loaded up ( can't see how it happens) or do you need to rescale the inputs so the desired comparative level sits at 0dB - I am trying to put in measurements taken in praxis
Thanks
Phil
Thanks
Phil
AFAIK there is v1.6 on the website, a newer version that ships with DLCPs and an ever newer version on beta test, so functionality is probably not finalized.
Software is a bit of a gray area at the moment and one of the two reasons (second being the lack of input/control boards) I keep postponing my purchase.
Software is a bit of a gray area at the moment and one of the two reasons (second being the lack of input/control boards) I keep postponing my purchase.