Have you considered the DspNexus?
dspNexus DSP Audio Processor - Danville Signal
Danville Signal Processing is a technology partner for high performance audio solutions. We provide SHARC-based DSP modules for a variety of OEMs, ODMs and oth
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
DEQX Premate 8 digital active crossover / DSP
- Thread starter Keith_W
- Start date
OP
Keith_W
Master Contributor
- Thread Starter
- #362
Guys, just because a box does DSP does not mean that the DSP is the same. The DEQX is a linear phase FIR DSP box. The only one in existence on the market. What disappoints me is that even in 2025 it is still the only one on the market. All the others - MiniDSP, Colinear, and Danville DSPNexus are all minimum-phase IIR. The only one I would avoid or caution about is the DSPNexus, because you program it with Audio Weaver. If you think DSP is difficult, wait till you see Audio Weaver. Add to that, Audio Weaver requires an annual subscription if you want to keep making filters. All other filter making software I can think of is a one-time purchase. Don't get me wrong, Audio Weaver is extremely powerful and flexible, and you can probably design any filter that you can think of. Read about something in an engineering journal, and you could probably implement it with Audio Weaver. But for the average consumer, you can forget it.
Regarding the "Audio Weaver" license on DspNexus: My understanding is that Danville Signal takes care of the cost, so it's zero dollars to the end-user.
Question: Is linear-phase FIR the "end-all, be-all", or are there alternatives? (l'm not an expert, so I'd like to know)
It would be nice to a see/hear a "live" shootout of the latest DSP offerings that are out there (e.g., DEQX-Pre8/4. DspNexus, MiniDSP).
OP
Keith_W
Master Contributor
- Thread Starter
- #364
AFAIK there is no perpetual license for Audio Weaver. You might want to check whether it's a one year license or a perpetual one.
Welllllllllllllll this is a bit of a can of worms. The alternatives are minimum-phase FIR and minimum-phase IIR. IIR can only be minimum-phase. There are methods to make a "quasi" linear phase IIR using subtractive filtering, but this is difficult to implement and requires you to input the equations into MATLAB or Octave yourself. I haven't tried it - but I have seen people do it.
Both FIR and IIR have advantages and disadvantages, but IMO the pros outweigh the cons for linear-phase FIR, which is why IMO it should be the filter of choice for audio applications. These are the advantages of linear-phase FIR over minphase IIR, in no particular order:
- ability to manipulate amplitude independent of phase. With minphase IIR, changing one will always change the other.
- because of the above, filter design is much simpler since you don't get phase cancellation. All XO filters of any Q and any order will always sum to flat.
- there is no phase distortion with linphase. Some people argue that phase distortion is inaudible, some say it is audible - see this thread.
- there are numerical advantages for FIR DSP since there is no feedback. In practice, processing is done with finite numerical resolution and this can introduce problems in an infinite loop feedback.
And these are the disadvantages:
- more computing power and more memory is required, particularly if you have a lot of taps
- there is both pre-ringing and post-ringing with linphase, as opposed to post-ringing alone with minphase. However, the pre and post ringing is theoretically eliminated with proper summation of HPF and LPF. In practice, it is never completely eliminated since drivers have different directivity, may not be aligned, etc. We try to keep pre-ring to a low level where it is inaudible due to the psychoacoustic phenomenon of pre-masking.
- there is a minimum latency for linphase FIR due to positioning of the impulse response in the middle of the filter length. This makes linphase FIR unsuitable for any application which requires ultra-low latency - e.g. studio monitoring and HT (where you get lip sync issues). HOWEVER, DEQX has a patent for a "zero latency" FIR.
Subjectively the audible difference is going to be quite small if both DEQX and MiniDSP are tuned by an expert. Both are capable of tuning the frequency response to below audible thresholds. The difference is that one introduces phase distortion, and the other doesn't. Whether you can hear phase distortion is controversial. I think you can, others think you can't. I have made linear-phase and minimum-phase versions of my filters. To me, linear phase sounds cleaner, more transparent, and transients have better attack. This was not a blind listening so make of that what you will.
The major difference is how tweakable linear-phase filtering is. The ability to adjust amplitude independent of phase opens the door to all sorts of fun experiments which I detailed in my system thread.
Thanks, Keith.
1) I will check with Danville about the license for Audio Weaver.
2) As I understand it, the latest version of DspNexus adds an FIR accelerator (compute outside of the DSP SHARK), so in theory FIR filters should be possible… But again, I will ask explicitly about linear phase filters.
This is the reply from Danville Signal regarding the Audio Weaver license:
“The license is free as part of the dspNexus unit. You will never pay for it.”
OP
Keith_W
Master Contributor
- Thread Starter
- #367
Good to hear that a perpetual license for Audio Weaver is included with the DSPNexus. I think that the only people who would be interested in a unit like that would be tinkerers - people who enjoy the process and want to play. If you are a "set and forget" kind of guy, MiniDSP is probably more suitable.
Also, FIR filters are not the same. Ask them how many taps are available. More taps = better. Read this thread for a more extensive primer on DSP. FYI, Acourate is capable of 262144 taps at 384kHz, and as many channels as you have CPU power available. I know a guy who has 30 channels of linear-phase FIR at 131k taps / 96kHz each. No hardware DSP is capable of anything close.
I’ve worked with the Danville Signal DSPmusik (an early Audioweaver DSP box) and still have it around. Unfortunately, the USB port is no longer functioning, so I can’t configure it anymore. That said, Audioweaver itself is quite interesting, it’s essentially a blank canvas where you build everything from scratch using blocks. The flexibility is amazing, but even something as “basic” as creating a three-way crossover with per-channel delay, EQ, and FIR corrections can be very time-consuming. It’s almost like soldering circuits together in software. Fun if you enjoy tinkering, but not plug-and-play. Also, its FIR capabilities are fairly limited and not the most stable.
I’ve also tried Audiolense, which offers full FIR filtering. The results were excellent, but the latency was a dealbreaker for me. Since I also use my Hi-Fi rig for movies, having hundreds of ms delay threw everything completely out of sync. Perhaps it’s possible to reduce that latency, but I didn’t explore further—mainly because the setup process was already very time-intensive and I wanted a solution that just always works by the push of a power button so my family is able to use it as well.
From what I’ve seen, DEQX and one other pro-audio company (the name escapes me) are the only ones that provide practical “zero-latency” FIR solutions. With DEQX, for example, the filtering always introduces a fixed 10 ms delay regardless of how complex your filters are. I don't know how DEQX handles this, that's there secret, but a clever way to reduce processing power and/or latency is through sample-rate conversion for the lower frequencies. Below 100 Hz, for instance, you don’t need 32-bit/192 kHz resolution. (even at 1khz you will never need this resolution) Dropping to something like 12-bit/22 kHz (or even lower) won’t affect sound quality at those frequencies, but it dramatically reduces processing load and when combined with the right set of filters with smaller FFT block sizes dramtically reduces latency as well. I’ve used this approach in Audioweaver often, and it makes a real difference by freeing up DSP resources for where they matter most. Believe me, nobody will ever hear the difference between 32bit/192khz vs 12bit/10khz like I used with AudioWeaver at anything below 100Hz. I have not tried any lower but I think it is possible to squeeze even more out of it. Just don't ever tell people who are listening to your system to get confirmation bias out of the equation.
Why do I choose FIR crossovers? As Keith explained, it’s mainly about reducing phase issues in the crossover region. With an FIR crossover, the filters can be designed to have a linear phase response. That means the drivers don’t add extra phase shifts from the filters themselves, so I only need to deal with the natural phase behavior of the drivers. This makes it much more predictable and therefore easier to achieve a clean, good-sounding result. It is possible with IIR filters but it takes a lot more effort to get the same results and in the end I'm also just a hobbyist.
I’ve also tried Audiolense, which offers full FIR filtering. The results were excellent, but the latency was a dealbreaker for me. Since I also use my Hi-Fi rig for movies, having hundreds of ms delay threw everything completely out of sync. Perhaps it’s possible to reduce that latency, but I didn’t explore further—mainly because the setup process was already very time-intensive and I wanted a solution that just always works by the push of a power button so my family is able to use it as well.
From what I’ve seen, DEQX and one other pro-audio company (the name escapes me) are the only ones that provide practical “zero-latency” FIR solutions. With DEQX, for example, the filtering always introduces a fixed 10 ms delay regardless of how complex your filters are. I don't know how DEQX handles this, that's there secret, but a clever way to reduce processing power and/or latency is through sample-rate conversion for the lower frequencies. Below 100 Hz, for instance, you don’t need 32-bit/192 kHz resolution. (even at 1khz you will never need this resolution) Dropping to something like 12-bit/22 kHz (or even lower) won’t affect sound quality at those frequencies, but it dramatically reduces processing load and when combined with the right set of filters with smaller FFT block sizes dramtically reduces latency as well. I’ve used this approach in Audioweaver often, and it makes a real difference by freeing up DSP resources for where they matter most. Believe me, nobody will ever hear the difference between 32bit/192khz vs 12bit/10khz like I used with AudioWeaver at anything below 100Hz. I have not tried any lower but I think it is possible to squeeze even more out of it. Just don't ever tell people who are listening to your system to get confirmation bias out of the equation.
Why do I choose FIR crossovers? As Keith explained, it’s mainly about reducing phase issues in the crossover region. With an FIR crossover, the filters can be designed to have a linear phase response. That means the drivers don’t add extra phase shifts from the filters themselves, so I only need to deal with the natural phase behavior of the drivers. This makes it much more predictable and therefore easier to achieve a clean, good-sounding result. It is possible with IIR filters but it takes a lot more effort to get the same results and in the end I'm also just a hobbyist.
@Keith_W: Ummh… Someone like you?
Well, order a DspNexus 2/8 and the license won't be an issue... And you could tell the "rest of us" how that box works!Exactly, someone like meI have been eyeing Audio Weaver for a while now, but the subscription thing puts me off.
(It sounds promising on paper... no pun intended
)This perspective is very helpful, thank you.
According to this post by someone using a unit:
.. the Colinear Acoustics DSP 8C allows linear phase FIR, even if the guy is having to import them from external FIR filter creation software (rePhase).
You may find many others can also be set up to use linear phase FIRs .. but simply that their user interface or software cant create them ? You'd have to investigate each one to discover if that might be the case ..
OP
Keith_W
Master Contributor
- Thread Starter
- #375
This is a common point of confusion. There is a difference between mixed phase DSP boxes (that Colinear, and also MiniDSP and DSPNexus) and a pure FIR box like the DEQX.
I will put it this way: with FIR, you can design any filter you can think of, including very high order filters. It will have superior numerical precision, it is always stable since it is not recursive, and it is capable of both linphase and minphase. However, you need A LOT OF TAPS, otherwise the filter will not have the characteristics that you want.
If you happen to be using a DSP box with insufficient taps - for e.g. some MiniDSP models have 4096 taps on both input channels at 96kHz (meaning 2048 taps per channel), the bin size is 96000/2048 = 46.9Hz. This is nowhere near enough to do anything meaningful, so the heavy lifting is done with IIR filters. The FIR is there for residual correction of the impulse response and it does not do much.
In contrast, the DEQX has an arm CPU with 32768 taps at 48kHz per channel - so the bin size is 48000/32768 = 1.47Hz. This means that all filters are FIR, there is no need for IIR. If you compare an 8ch DEQX with an 8ch MiniDSP, the DEQX has 262,144 taps in total, and the MiniDSP is 4096. I can't tell you how many taps the Colinear has, because I haven't looked into it in any detail. But it will be very similar to the MiniDSP since it uses the same DSP chips.
If you would like to understand a bit more about the difference between FIR and IIR, please read this thread.
Sadlly, all of the Internet searches I’ve made have uncovered a very unhappy story to this point in the development of the DEQX Gen4… I hope that it has a happy ending, but it’s not trending in that direction!
Can you specify what the point of common confusion is, just to be clearer.
I *presume* - but a presumption is still uncertain and guessing, reading between lines - you are saying it's a common confusion that linear phase FIRs can be created with less than xxxxx number of taps per channel available, yet you don't say how many taps you think are needed or for what frequencies you're talking about, nor what shape filters youd be applying. .
You appeared to be asserting that the DEQX gen 4 is the only box that can do linear phase FIRs .. *at all* .
But surely that's not true nor relevent for crossovers in a 3 way xover design where you're working at, say, 3k and 500 Hz points and say 48db/Oct slopes or shallower. How many taps are needed in that scenario?
This is all going to be highly specific as to what you're trying to do with the DSP .. and not an either/or whether any unit can do linear phase FIR at all.
Anyway .. the DEQX premate 8 is £13,000 over here. You could buy 8 Colinear DSP boxes 49000 taps per box.. and still be £5k better off. Or a secondhand Trinnov for £3k which will also be usimg a generic processor for many more taps, like the DEQX has , if not more still (there's nothing pure FIR about an ARM processor).
OP
Keith_W
Master Contributor
- Thread Starter
- #378
What I meant is - a DSP box with a few FIR taps (4096 in the case of MiniDSP) is not the same as a DSP box with many FIR taps (32768 for DEQX). Maybe this illustration will help:
Here, we have a subwoofer measurement. In red, we have 131,072 taps @ 48kHz (I increased the gain a couple of dB so that it's not hidden under the other measurements). In green, it has been cut down to 32,768 taps to simulate a DEQX. You can see that the shape is an almost exact copy of the high-tap measurement. In blue, the measurement has been cut down to 1024 taps @48kHz (this will be the same as 2048 taps @ 96kHz, which is what we have in the MiniDSP).
And here is what happens if you try to replicate a steep filter with a low tap count FIR filter. In green, we have an 8th order LR @ 50Hz with 32768 taps. In blue, the same filter but cut down to 1024 taps.
This is why I said that in a box like MiniDSP / Colinear, the heavy lifting is done with IIR filters. There aren't enough FIR filter taps to make high order linear-phase filters, or high resolution corrections. The minimum that you need is probably 16384 - 32768 taps @ 48kHz.
Not disputing that. See my first post in this thread. IMO DEQX has priced themselves out of the market. I can speculate why they made that business decision - maybe their manufacturing and support capacity is limited, and having too many customers means they can't make enough units and support the customers that they do have. Remember that businesses and consumers have different priorities. Their priority is to make money and stay in business. If they can sell enough units to do that, then it is a sound business strategy. But as a consumer, I agree that it sucks. If it was priced the same as a MiniDSP and they suddenly have to manufacture and sell thousands of units and service thousands of consumers ... that would be a disaster, too.
Look over the pond and see what MiniDSP are doing. They have extensive documentation, easy-to-follow guides, and a support forum with some staff, but mostly experienced community members who lend free time to help others. That kind of thing does not appear overnight. What MiniDSP has is the result of a different business strategy with many years of well-planned growth and a lot of consumer goodwill. This is why they are the 800lb gorilla in the room, it is very difficult to compete against them. DEQX OTOH are small, do not have a community forum, has virtually no documentation available to the public, and support is provided one-on-one.
I am not sure what you mean about "nothing pure FIR about an ARM processor". An ARM CPU is a general purpose CPU. What is important is that it has enough grunt to run FIR with a high tap count. If you wanted to, you could make it run mixed phase. I haven't looked too closely at Trinnov, but my understanding is that there is a PC in there with a whole bunch of custom cards. I don't know if it's full FIR or mixed phase. Given that it's an AVR, I would guess that they would prefer mixed phase for the low latency.
MiniDSP is an excellent product, I would happily recommend it to anyone. With DEQX, let's just say it has its pros and cons. The DSP is theoretically better, which I hope I have adequately explained. But ... there are other considerations too. Like price and support.
Let med share a different perspective on this. I'm not an expert on FIR filters myself, but have contacts who are. And I have tested things in practice and done AB comparisons.
Firstly, there's not considered by experts in the field to be an audible benefit in running anything higher than a 6th order FIR for crossover. Which means we don't need that many taps. The Colinear DSP-8C will have more than what's needed and one could also use 8th order with high resolution with it, though it's not considered to offer any audible improvement.
Secondly, the FIR taps in miniDSP Flex (2048) is not useless at all. With a well designed FIR filter, the improvement over IIR is easily and clearly audible. I have done the comparison. Having a bit more taps than the Flex would be better in theory, but it's not definitive that will be audible. This may also depend on several things.
What seems useless IMO is the number of taps Hypex plate amplifiers have, which is only 1024 taps per channel. I have not been able to get the good results for a FIR crossover with that few taps. But 2048 taps works really well.
While we think that DEQX has priced themselves out of the market, they are thinking the opposite. They have priced them into the high-end market when such a cost is very much needed to sell to it. We were planning to offer the DEQX with some of our speakers when the price was going to be what the former DEQXs units cost. Now with the new price tag, I'm struggling with a good conscience to offer it.
Firstly, there's not considered by experts in the field to be an audible benefit in running anything higher than a 6th order FIR for crossover. Which means we don't need that many taps. The Colinear DSP-8C will have more than what's needed and one could also use 8th order with high resolution with it, though it's not considered to offer any audible improvement.
Secondly, the FIR taps in miniDSP Flex (2048) is not useless at all. With a well designed FIR filter, the improvement over IIR is easily and clearly audible. I have done the comparison. Having a bit more taps than the Flex would be better in theory, but it's not definitive that will be audible. This may also depend on several things.
What seems useless IMO is the number of taps Hypex plate amplifiers have, which is only 1024 taps per channel. I have not been able to get the good results for a FIR crossover with that few taps. But 2048 taps works really well.
While we think that DEQX has priced themselves out of the market, they are thinking the opposite. They have priced them into the high-end market when such a cost is very much needed to sell to it. We were planning to offer the DEQX with some of our speakers when the price was going to be what the former DEQXs units cost. Now with the new price tag, I'm struggling with a good conscience to offer it.
So .. as we can summise from your explanations of resources needed for FIRs, the statement in the quote above is just hyperbole.
There is no such thing as a "linear phase FIR DSP box" for starters and you're claim that it is a "pure FIR box" is an equally odd statement to make.
t uses, as you agree and acknowledge, a generic CPU which can be, and has been, utilised to allow more than FIR filters - where does this "purity" factor in?It's not a specialised chip for FIR creation as that would imply, the opposite in fact ( some of the units using specialised DSP chips could be said to be more 'pure' in that sense .. just that their power is lacking compared to generic processing).
And ... let's turn to the elephant in the room here that looms over your posts mentioning DEQX, above : The DEQX gen-4 with 32,00+ taps per channel ... (drum roll)
*does not exist*
lol.
So, to say it is the "only [linear phase FIR DSP box] on the market " is strange rhetoric considering it is not on the market AND when all the other DSP offering FIR filters *CAN* do linear phase FIR ..
You need to specify that what you actually mean is *meaningful linear phase FIRs for sub integration"
When one is not using a sub and not requiring super steep filtering, .. many of the DSP boxes have a fair amount of taps available for FIR which is why the guy in the other thread *IS* using the Colinear for linear phase FIRs and I do not think is confused about it .. as you think he must be.
I just wanted to guide the conversation away from generalised hyperbole (when it's quite a specific scenario this is relevant for .. if audible anyway) and bring it back to reality of what is needed in the real world .
Similar threads
