dspNexus DSP Audio Processor - DANVILLE SIGNAL - SHIPPING NOW?

[Keith_W]

I was not aware of that subtractive filter method, thank you. I will read that DIYAudio thread you linked and experiment a bit with it.

I suppose the use case of the DSPNexus is for a more advanced DSP user than MiniDSP. Audioweaver may be complex, but it is also very flexible. I would be very impressed if you can use it to coax a linear phase filter from a DSPNexus. Can't say the same about MiniDSP. I borrowed a MiniDSP for a few weeks and I was impressed by how simple it was to use. But ultimately I found it to be quite limiting.

FWIW, I am not opposed to DSPNexus as a concept, I just didn't know how it fits in the marketplace. Now I know. Maybe i'll buy a unit and play with it a bit. I know FIR filters, but I know a lot less about IIR's. They just seem so difficult to design. Certainly being able to achieve linear phase to 20kHz with no pre-ringing and minimal latency sounds very attractive. Also attractive is the challenge of learning to use AudioWeaver ...

 

[3ll3d00d]

Fwiw I added a subtractive filter option to https://beqdesigner.readthedocs.io/en/latest/ui/manage_mc/ so you can use that to design such a thing for multiple ways (and load it into jriver automatically).

for example, this is pulled from my live tv viewing dsp, most of that mess is the 4 way subtractive filter

the dip in the sum is because there's an additional time alignment offset baked into this, the raw filter would not show such a dip but the physical speaker has that offset

anyway, attempting to create such a filter by hand is (IMO) a total nightmare and I couldn't find anything to calculate it so I wrote one myself. I guess you can use the same to transcribe to some other format though.

 

[gnarly]

I am sorry, but this is incorrect. You can't have linear phase with IIR, because linear phase requires symmetrical mirror image zeros outside the Z-plane unit circle. That is not possible with IIR, all the zeros have to be within the unit circle. It is possible to have linear phase regions of the frequency response, but all XO's and all PEQ's are minimum phase.

Technically, I think you are correct. Linear phase is commonly associated with a symmetrical filter around impulse peak.
I think where it all gets a bit confusing, is when we sometimes erroneously automatically equate flat phase at zero as being linear phase.

Techniques to achieve flat phase at zero, like @mdsimon2 shows and links, essentially trade pieces of fixed time delays in a manner to best fit reducing phase rotations.
Imo, the all-pass technique is so impractical, it's a non starter.

I was a participant in the linked APF thread. The gent who gave the example for linearizing phase on a summed LR 24 dB/oct crossover at 1kHz
(excuse me, I should say flattening phase example ) ,
was very nice to provide the series of all-pass he used. It took 13 APF's to achieve this:

Ignore the black vertical phase wraps; they are the phase of the individual APF's that I can't remove from display.
The wavy phase trace in red and green is the summed result. Notice the phase scales so waves aren't as bad as looks.
But look at the impulse response. The oscillations before impulse peak are due to needing even more all-pass above the last one used at about 19kHz.
(there was 100ms of delay added to both channels for ease of display; true impulse peak would be around 0.78ms)

Ok, all that is what 13 APF biquads and the additional biquads needed for the LR24 hp and low pass will get you.
With 0.78ms fixed delay.

Here's how two channels of linear phase FIR, only 128 taps, can make the same LR 24 dB/oct at 1kHz. (FIR @ 48kHz)
0.42ms fixed delay.

Personally, based on this delve into all-pass, I think saying APF's can be used to flatten phase is technically correct, but pragmatically unreal..to the point of we probably shouldn't even say it can be done...other than academically.

I've played with comparing FIR with subtractive filters in other threads on DIY, and have the same opinion.......WHY?

 

[gnarly]

Fwiw I added a subtractive filter option to https://beqdesigner.readthedocs.io/en/latest/ui/manage_mc/ so you can use that to design such a thing for multiple ways (and load it into jriver automatically).

for example, this is pulled from my live tv viewing dsp, most of that mess is the 4 way subtractive filter

View attachment 466552

the dip in the sum is because there's an additional time alignment offset baked into this, the raw filter would not show such a dip but the physical speaker has that offset

View attachment 466549

anyway, attempting to create such a filter by hand is (IMO) a total nightmare and I couldn't find anything to calculate it so I wrote one myself. I guess you can use the same to transcribe to some other format though.

Very cool...i feel your pain !!!!

Here's the essential processing blocks needed for 5 way linear phase, other than I/O routing etc.

To get this measured 5-way acoustic results. To heck with IIR...well, not quite ... I do use IIR system high-pass.

Btw, processor is used Q-Sys Core510i, that routinely sell on ebay for $1000-2000 depending. I've run 15 channels of 16k taps with it. (48kHz only)

 

[mdsimon2]

@gnarly

I agree with your take on phase flattening with APFs, that seems completely impractical.

I can see the case for a linear phase subtractive crossover in a processor with limited processing resources (i.e. no FIR). Okto's XMOS DSP is a perfect example of this, no FIR capability and IIR resources are limited, but you can easily implement a subtractive filter because the design is flexible.

Michael

 

[gnarly]

@gnarly

I agree with your take on phase flattening with APFs, that seems completely impractical.

I can see the case for a linear phase subtractive crossover in a processor with limited processing resources (i.e. no FIR). Okto's XMOS DSP is a perfect example of this, no FIR capability and IIR resources are limited, but you can easily implement a subtractive filter because the design is flexible.

Michael

Sure. That makes good sense when we have something we like, and can make it work.