Jump to content

Fir filter thoughts?


NBPK402

Recommended Posts

  • 5 months later...

Sorry I missed this thread, Ron.  I think that the video is sort of a "me too (but I'm not going to tell you that the technology has been around for quite a while)". 

 

I.  What good is "FIR" filtering over what I currently have?

 

FIR filtering is a way to correct not only the SPL response (a.k.a., "frequency response") of your loudspeakers in-room, but also their phase response.

 

II. "So What?"

 

Well, a lot...actually.  Experiences using "linear phase loudspeakers" have been documented pretty widely since the early 2010s in the home hi-fi marketplace.  Here is a pretty good article: http://www.linkwitzlab.com/Attributes_Of_Linear_Phase_Loudspeakers.pdf

 

Quote

"...Another area in which loudspeakers are disreputable is in the neglect of the time domain. The traditional view is that all that matters is to be able to reproduce continuous sine waves over the range of human hearing. A very small amount of research and thought will reveal that this is a misguided view. Frequency response is important, but not so important that the attainment of an ideal response should be to the detriment of realism. One tires of hearing that 'phase doesn't matter' in audio, or 'the ear is phase deaf'. These are outmoded views which were reached long ago in flawed experiments and which are at variance with the results of recent psychoacoustic research."

 

III. So What is the Difference in Flat or Linear Phase Response from What I've Got Now?"

  1. Tighter and deeper perception of bass
    "Phase equalization of the bass...subjectively extends the effective bass response by the order of half an octave...
     
  2. Wider and deeper sound stage (quite dramatic, in fact)
    "Without [flat phase response], the sound [is] flat and almost lifeless in comparison."
     
  3. Greater realism
    "...the initial [sound] transient and [its] relaxation time are critical for realism. Anything in a sound reproduction system which corrupts the initial transient is detrimental [to the perception of realism]."
     
  4. Apparent soundstage depth
    "This may surprise some listeners when they first hear it, since many speakers (and records) elicit only a general left-to-right spread. But "stereo", as originally conceived, implied a three-dimensional sound in which voices or instruments could be localized at different apparent distances from the listener as well as at various lateral positions. Listeners to time-aligned speakers consistently report hearing a stereo image with unusual depth."
     
  5. Greater Resolution
    "The stereo image is reproduced precisely, each voice or instrument having its proper place and width. In complex sound sources such as symphony orchestra, individual instruments can be resolved with unexpected clarity. In the old cliche, "I hear details I never knew were in the recording. " Some listeners have incorrectly attributed the improved resolution of detail to more accurate transient response, but the better definition of details is simply the result of the reduction of blending in the stereo image."
     
  6. Separation of ambience
    "With loudspeakers whose stereo image is slightly blended because of time-smear, any hall ambience or reverberation in the recording tends to become slightly mixed with the instrumental sounds, causing coloration of those sounds. Consequently, with such speakers closely microphoned recordings tend to sound better because of their distinctly defined sound. But with time-corrected loudspeakers, the ambience is resolved as a separate sound, and larger amounts of hall ambience in recordings can be enjoyed...”

 

IV. So why haven't more home hi-fi loudspeakers incorporated FIR filtering?

 

This is perhaps the most interesting part of this subject.  Basically, I'd characterize it as anti-digital bias by what I term "mossback audiophiles"...because FIR filtering is a type of digital filtering. It really doesn't exist in the "analog" world.  It's the mossback audiophiles that grew up in the 1950s-1970s that have rejected DSP loudspeaker correction...but curiously, get all of their new music digitally (whether they realize it or not).

 

Chris

  • Like 1
Link to comment
Share on other sites

Tried to play little bit with FIRs and my crossover (Najda DSP). Unfortunately low frequency filtering has huge processing requirements. My XO allows only 1024 tapes per channel, mini dsp 2048. Good starting point would be 10k and more for sub. Anyway FIR's allow for very steep XO points. -90dBa is not a big deal.

  • Thanks 1
Link to comment
Share on other sites

Another thing I like about FIR filters that they don't change phase at crossover point. So signals from two drivers are always adding or subtracting (depending on a choice) and we avoid those strange wobbles near XO point where phase changes several times.

Link to comment
Share on other sites

On 4/19/2021 at 3:22 AM, Bacek said:

Anyway FIR's allow for very steep XO points. -90dBa is not a big deal.

 

There is a fundamental relationship between the steepness of the filter slope and the ringing in the time domain. Filters this steep ring. So this is always a tradeoff.

  • Like 1
Link to comment
Share on other sites

3 hours ago, Bacek said:

Another thing I like about FIR filters that they don't change phase at crossover point. So signals from two drivers are always adding or subtracting (depending on a choice) and we avoid those strange wobbles near XO point where phase changes several times.

 

To be clear, the statements above are only true if the filter is designed for them. You can design a FIR filter that emulates an IIR filter, if you want to. These things are all part of the design specs and tradeoffs.

  • Like 1
Link to comment
Share on other sites

2 hours ago, Edgar said:

From the obscure archives: Transversal filtering of analog signals with a tapped analog delay line

It never really caught on for general consumption.

Analog delay lines are not very "hi-fi", the last time I looked.  I was aware of this, but I think my comment still stands: FIR filtering really only exists in the digital world.

 

2 hours ago, Edgar said:

There is a fundamental relationship between the steepness of the filter slope and the ringing in the time domain.

There are a lot of comments almost everywhere about "measures great but the results suck the life out of the music", so I think a greater knowledge of the way that the human hearing system works is probably warranted in this "new" area.  I would think this should capture a lot of attention by our more serious audio enthusiasts that are just now starting out their audio journeys...

  • SPL response flatness:  I've found that a sort of "economy of filtering" works, but one with an exception:  flatter SPL response.  I've found that the flatter the SPL response, the better the sound. 

    When combined with somewhat flattened phase response, and most importantly--control of early reflections in-room from the loudspeakers to the listener's ears via loudspeaker directivity and proper placement of nearfield midrange absorption--the results are startling.  I'm also talking about directivity control below 500 Hz--and that really implies horn-loaded bass with directivity control down to at least the room's transition frequency (i.e., 100-200 Hz for typical home-sized listening rooms).
     
  • Phase response: I've found that swings of ±90 degrees of phase vs. frequency...are not really very audible.  In other words, it may be that ±90 degrees might be "good enough".  This is a subject that I believe warrants further attention, because the tradeoff is time delay and added computational requirements.  This also corresponds to Danley's comments about getting acoustic drivers to within a quarter wavelength at the crossover frequency.  This is the principle of the multiple-entry horn (MEH), and one that I find works for subwoofer placement, too (generally at a frequency below the room's Schroeder or transition frequency). 

 

So some present "rules of thumb" that I think apply in this subject area, and other principles are just now coming into better focus.  Applying IIR filters to do the heavy lifting first using the PEQ filters and time delays in typical DSP crossovers that might not have FIR filtering capabilities--or sufficient FIR filtering capacity in terms of the number of taps per channel, then applying an upstream secondary FIR filtering capability (perhaps using JRiver, etc. and rePhase) seems to be the current trend.  YMMV.

 

I do know one thing: there still is an advantage that horn-loading seems to bring to the table, and that turns into the listener being able to hear the differences between flat SPL and phase response without having to put their loudspeakers out in the middle of their listening room and thereby take a severe hit on the resulting increased bass modulation distortion that comes with this practice.  You really can have the best of both worlds if using fully horn-loaded loudspeakers.: using the room corners for much better bass, and flattened SPL/phase using DSP crossovers and upstream FIR filtering to flatten phase response further.

 

Chris

Link to comment
Share on other sites

So, I see in the vid above, the piece of gear pointed to is the Marani LPP 480, which is $1600 bucks on eBay. anything else out there that does the same thing for less?     

https://www.ebay.com/itm/Marani-LPP480A-FIR-4IN-8OUT-Speaker-Mgmt-System-ONLY-AUTHORIZED-USA-DEALER/164087719843?hash=item263463e7a3:g:854AAOSw43haKHnt.

 

 

Link to comment
Share on other sites

This is much more affordable: https://wiki.jriver.com/index.php/Convolution, especially if you already own a capable PC or laptop.  This approach overlays a standard non-FIR DSP crossover with an upstream PC (as I mentioned above).

 

Another thread that explores this subject using a miniDSP OpenDRC-DA8: https://www.audiosciencereview.com/forum/index.php?threads/multi-channel-multi-amplifier-audio-system-using-software-crossover-and-multichannel-dac.12489/page-8

 

It appears that the OpenDSC-DA8 doesn't have nearly enough available taps per channel.  That's apparently not an issue with JRiver.

 

Chris

 

  • Thanks 1
Link to comment
Share on other sites

8 hours ago, Chris A said:

Analog delay lines are not very "hi-fi", the last time I looked. 

 

No, but filters are used in more places than audio. 😉

 

Quote

There are a lot of comments almost everywhere about "measures great but the results suck the life out of the music", so I think a greater knowledge of the way that the human hearing system works is probably warranted in this "new" area.

 

There is quite a lot of available knowledge about this in psychoacoustics. For linear-phase filters in particular, it is known that "pre-ring" (ringing that begins before the event) is far more audible and objectionable than "post-ring" (ringing that occurs after the event). Linear-phase filters, by definition, have both pre-ring and post-ring, and they are in fact identical but reversed in time. Minimum-phase filters have only post-ring.

 

Also, if the anomaly being equalized is itself minimum-phase, then the equalization to compensate for it is also minimum-phase. Linear-phase filters in that application are inappropriate.

 

Quote

Phase response: I've found that swings of ±90 degrees of phase vs. frequency...are not really very audible.  In other words, it may be that ±90 degrees might be "good enough".

 

Thanks for mentioning that, Chris. It's something that I've wondered about, but I do not have the facilities to experiment myself. It does make the problem easier to solve.

 

I should also mention that signals within 90° phase add to within 3dB of the amplitude they would have if they were perfectly in-phase.

 

Quote

Applying IIR filters to do the heavy lifting first using the PEQ filters and time delays in typical DSP crossovers that might not have FIR filtering capabilities--or sufficient FIR filtering capacity in terms of the number of taps per channel, then applying an upstream secondary FIR filtering capability ... seems to be the current trend.

 

Beyond a few tens or hundreds of FIR taps, transform-domain filtering (FFT, filter, IFFT) becomes more computationally efficient. 

 

It should also be noted that a necessary part of linear-phase filtering is time delay. This is not a problem if only listening to music, but if the audio has to be synchronized with video then it very quickly becomes a problem. A 30 Hz video frame spans 1470 samples at 44100 Hz, so any linear-phase filter with more than 2940 taps is automatically at least a full frame out of sync. Add to that the buffering required by transform-domain processing, and the latency can easily exceed several video frames.

 

One final comment: There is a tendency to equate FIR filters and linear-phase filters. Linear-phase filters must be FIR [1], but not all FIR filters are linear-phase. They have to be designed for it.

 

- Greg

 

[1] IIR filters can be constructed that approximate linear phase within a desired band.

Link to comment
Share on other sites

49 minutes ago, Edgar said:

Beyond a few tens or hundreds of FIR taps, transform-domain filtering (FFT, filter, IFFT) becomes more computationally efficient. 

Perhaps there needs to be more discussion on that subject for hi-fi audio applications, which I haven't really seen much of in the context of hi-fi audio...speaking frankly.

 

I actually see a lot of demand materializing for flat-phase bass in loudspeakers, but no real way to get there without using thousands of taps at the resolution needed for hi-fi audio (roughly the 15 Hz-->20 kHz pass band). 

 

Chris

Link to comment
Share on other sites

21 minutes ago, Chris A said:

Perhaps there needs to be more discussion on that subject for hi-fi audio applications, which I haven't really seen much of in the context of hi-fi audio...speaking frankly.

 

BruteFIR is gaining a lot of acceptance in this role. https://torger.se/anders/brutefir.html#whatis Partitioned convolution -- removes much of the buffering latency of large FFT/IFFT filtering, but nothing can remove the inherent delay in linear-phase filters.

Link to comment
Share on other sites

Reading through that BruteFIR documentation takes me back to the days of spending a great deal of time on Sun Workstations in the mid-late 1990s.  I know the benefits of Linux in terms of preserving full CPU/memory horsepower of PC hardware, but you have to admit--this is a pretty deep dive for anyone but those that are very familiar with Linux and PC hardware nowadays, with significant computer engineering background, i.e., it's not anything close to "plug 'n play". The last time I was into Linux was mid-1990s...when the OS was still in its infancy.  But it looks like it can fully support a 5.1 array, depending on length of the FFTs and the number of channels. 

 

Chris

Link to comment
Share on other sites

16 minutes ago, Chris A said:

... you have to admit--this is a pretty deep dive for anyone but those that are very familiar with Linux and PC hardware nowadays, with significant computer engineering background, i.e., it's not anything close to "plug 'n play".

 

While I am perfectly willing to give BruteFIR the signal processing credit that it deserves, you and I are of like minds with regard to Linux. I abhor Linux with a passion that I find difficult to put into words ... and that goes double for Linux Audio. My loathing comes from direct experience, not from prejudice.

Link to comment
Share on other sites

Here's the problem with Linux:

 

Quote

[Linux] Memory management

See also: Memory management

Memory management in Linux is a complex topic. First of all, the kernel is not pageable (i.e., it is always resident in physical memory and cannot be swapped to the disk). In the kernel there is no memory protection (no SIGSEGV signals, unlike in userspace), therefore memory violations lead to instability and system crashes.[176]

This is a big problem as a multi-tasking OS, especially a multi-user OS.  I see they still haven't done any OS development to deal with these issues.  Real OSes in my terminology are either 1) RTOS's or, 2) bullet-proof multitasking/multi-user with memory protection.  Linux is neither.  It's still just a "me, too" PC operating system that allows the user full control of the OS and hardware support (and YOU must support it--in everything it does--even going as far as compiling and linking it).  This isn't good computer science--in a nutshell.

 

The command line syntax is basically Unix syntax (with very few differences, the last time I looked--25 years ago).  I don't know what they're using for a windowing system nowadays, but there was a big push in the mid-90s to use X-Windows, which was okay.  I'm sure that it's changed since then.

Link to comment
Share on other sites

On 4/18/2021 at 2:45 PM, Chris A said:

 

IV. So why haven't more home hi-fi loudspeakers incorporated FIR filtering?

 

This is perhaps the most interesting part of this subject.  Basically, I'd characterize it as anti-digital bias by what I term "mossback audiophiles"...because FIR filtering is a type of digital filtering. It really doesn't exist in the "analog" world.  It's the mossback audiophiles that grew up in the 1950s-1970s that have rejected DSP loudspeaker correction...but curiously, get all of their new music digitally (whether they realize it or not).

 

Chris

 

Haha, I like this !!    I'd call myself a mossback who's seen the light 😀

Been a heavy FIR user for about 6 years now.

Link to comment
Share on other sites

On 4/21/2021 at 7:36 AM, Edgar said:

 

There is a fundamental relationship between the steepness of the filter slope and the ringing in the time domain. Filters this steep ring. So this is always a tradeoff.

 

Hi!   True, for one sided filters, high-pass or low-pass. 

But my understanding is not true for complementary linear phase xovers, where both pre and post ringing cancel each other out.

I feel complementary linear phase xovers are probably the most overlooked no-brainer way to increase SQ out there. (along with concomitant mutli-amping etc)

 

Of course it's easy enough to achieve fully complementary xovers electrically, but quite a bit to harder to achieve them acoustically. 

Since a speaker needs to be acoustic complementary to avoid ringing, I'm an advocate of getting the drivers as complementary as possible pre-xover.

By using minimum phase EQs driver-by-driver to flatten mag and phase within the drivers' pass-bands like as always done, but also continuing such flattening through the drivers' xover summation regions. 

Then just add complementary linear phase xovers to tie the drivers together.

I've found steep xovers minimize the out of band min-phase mag flattening needed, which is always the most difficult task. And it 'undoes' nearly all the out-of-band mag boost that flattening required , which could otherwise harm drivers.

Steep also reduces the width of the xovers summation, reducing the freq range of off-axis lobings.

 

And bingo...relatively (if not completely)  flat phase, with no ringy dingy !!

Anyway, my 2c thoughts 😃

Edited by gnarly
grammar
Link to comment
Share on other sites

4 minutes ago, gnarly said:

But my understanding is not true for complementary linear phase xovers, where both pre and post ringing cancel each other out.

 

True, they cancel each other out mathematically, but as you pointed-out, getting them to cancel acoustically is a huge problem in itself, even with identical drivers. The ringing in one driver is identical to the complementary ringing in the other, but of opposite polarity. On-axis, the two cancel each other out. Off-axis, they do not. That energy still goes into the room, different portions of it at different angles relative to the main axis of the loudspeaker. It is uncontrolled energy that corrupts the signal.

 

 

Link to comment
Share on other sites

18 hours ago, Edgar said:

It should also be noted that a necessary part of linear-phase filtering is time delay. This is not a problem if only listening to music, but if the audio has to be synchronized with video then it very quickly becomes a problem. A 30 Hz video frame spans 1470 samples at 44100 Hz, so any linear-phase filter with more than 2940 taps is automatically at least a full frame out of sync. Add to that the buffering required by transform-domain processing, and the latency can easily exceed several video frames.

   Thanks for your insight into linear phase filters, it is a fascinating subject.  Could this problem be solved by also delaying the video signal?  It wouldn't work for video games, obviously, but it should work for movies/video, right?

    What hardware are people using to incorporate FIR filters into their hi-fi system?

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...