John Atkinson: Yes, MQA IS Elegant...

[mansr]

Here's an example impulse response of a linear phase filter creating a sharp 6 dB dip between 1 kHz and 2 kHz at a sample rate of 48 kHz:

The time axis is marked in seconds.

 

Here's the frequency response:

 

The peak amplitude of the ripple is about -34 dB. That means the "ringing" will be 34 dB below whatever content the signal has at those frequencies. The zip file attached to this post contains two files. One consists of this impulse response repeated 10 times, the other the equivalent minimum phase response. See if you can hear any difference.

 

This is for illustration only. A realistic correction filter wouldn't have such steep transitions and thus less "ringing."

p.zip

[Shadders]

5 minutes ago, Hifi Bob said:

Not so—have a read of http://www.dspguide.com/ch17.htm

Hi,

In terms of a rooms equalisation, where there are many peaks and troughs, and as per the text in their website - they use multiple filters.

Regards,

Shadders.

[Jud]

13 minutes ago, Hifi Bob said:

Not so—have a read of http://www.dspguide.com/ch17.htm

 

Didn't see anything I recognized immediately as relevant to room correction there. Could you explain further?

[Jud]

@mansr - possible to also provide two files, one with min phase left channel, linear phase right channel simultaneously, and the other with channels reversed? 

[mansr]

3 minutes ago, Jud said:

@mansr - possible to also provide two files, one with min phase left channel, linear phase right channel simultaneously, and the other with channels reversed? 

Here you go.

ps.zip

[Shadders]

1 hour ago, Jud said:

 

Didn't see anything I recognized immediately as relevant to room correction there. Could you explain further?

Hi Jud,

I had a quick read - they are using the proposed frequency response to generate a single filter. The response is likely to be the inverse of the room measurement.

 

From the matlab wedsite - seems to be the same example :

https://uk.mathworks.com/help/dsp/examples/arbitrary-magnitude-filter-design.html

FIR Modeling with the Frequency Sampling Method

This section illustrates a case where the amplitude of the filter is defined over the complete Nyquist range (there are no relaxed or "don't care" regions). The example that follows uses a single (full) band specification type and the robust frequency sampling algorithm to design a filter whose amplitude is defined over three sections: a sinusoidal section, a piecewise linear section and a quadratic section. It is necessary to select a large filter order because the shape of the filter is quite complicated:

 

The REW system etc., allows you to remove sections of the equalisation, such as bass frequencies that have been amplified (neighbours), and implement your required granularity for the spectrum modifications.

 

Regards,

Shadders.

[mansr]

1 hour ago, Shadders said:

I had a quick read - they are using the proposed frequency response to generate a single filter. The response is likely to be the inverse of the room measurement.

Simply put, to create a room compensation filter, you take the measure response, invert it, and compute the inverse Fourier transform. This gives you the impulse response.

[crenca]

1 minute ago, mansr said:

Simply put, to create a room compensation filter, you take the measure response, invert it, and compute the inverse Fourier transform. This gives you the impulse response.

 

I'm with Jud in that there appears to be more than one (I have counted 3 so far) definitions of "phase" and "impulse response" being used on the last couple of pages of this thread.  Perhaps "definition" is not the right word, but it is confusing when the same terms get applied to instances that our so different.  The math is the unifying factor I presume...

 

Also, have we not come to the conclusion that the impulse that JA used to illicit "ringing" in his article is in fact an "illegal" frequency outside of the 20-20 band limit?  This the the use case of "impulse response" as related to "ringing" and filter design and thus "phase" of said filters, not convolution related to room correction no?

[Jud]

2 hours ago, Shadders said:

Hi Jud,

I had a quick read - they are using the proposed frequency response to generate a single filter. The response is likely to be the inverse of the room measurement.

 

From the matlab wedsite - seems to be the same example :

https://uk.mathworks.com/help/dsp/examples/arbitrary-magnitude-filter-design.html

FIR Modeling with the Frequency Sampling Method

This section illustrates a case where the amplitude of the filter is defined over the complete Nyquist range (there are no relaxed or "don't care" regions). The example that follows uses a single (full) band specification type and the robust frequency sampling algorithm to design a filter whose amplitude is defined over three sections: a sinusoidal section, a piecewise linear section and a quadratic section. It is necessary to select a large filter order because the shape of the filter is quite complicated:

 

The REW system etc., allows you to remove sections of the equalisation, such as bass frequencies that have been amplified (neighbours), and implement your required granularity for the spectrum modifications.

 

Regards,

Shadders.

 

My error.  I was automatically thinking of room correction filters as dealing with separate frequency ranges, but the implementation can be done with a single filter.

[Le Concombre Masqué]

5 hours ago, mansr said:

That doesn't make much sense. The output of an IIR filter is a linear combination of input samples and previous output samples. FIR filters use only input samples to create the output, making them a subset of IIR filters. If anything IIR filters are more "powerful" since they can achieve things impossible with FIR. For example, an impulse response consisting of a step function is trivial as an IIR (current input + previous output). IIR filters can also be unstable and oscillate or increase without bound. Some FIR filters have an equivalent IIR filter with fewer taps, which is cheaper to realise. This probably what they are referring to in the last bit.

what do you think of Cascaded integrator–comb filter ? doesn't moving average promoted by Ayre belong to that family?

[mansr]

2 hours ago, crenca said:

I'm with Jud in that there appears to be more than one (I have counted 3 so far) definitions of "phase" and "impulse response" being used on the last couple of pages of this thread.  Perhaps "definition" is not the right word, but it is confusing when the same terms get applied to instances that our so different.  The math is the unifying factor I presume...

A filter is defined by its impulse response. The frequency response of a filter is the Fourier transform of the impulse response. For each frequency, this yields a magnitude and a phase. With a linear phase filter, the phase response is proportional to the frequency. This means relative alignment of frequency components is preserved since the time delay is constant. Minimum phase filters do not have this property, delaying frequencies by varying amounts. These filters are favoured for live effects as the overall latency is small and phase distortion is mostly inaudible.

 

Quote

Also, have we not come to the conclusion that the impulse that JA used to illicit "ringing" in his article is in fact an "illegal" frequency outside of the 20-20 band limit?

He used an illicit pulse to elicit ringing. He also sent the signal through a multitude of DACs and ADCs for no apparent reason. It is unclear what he expected to prove and murkier still what his experiment actually shows, if anything.

 

Quote

This the the use case of "impulse response" as related to "ringing" and filter design and thus "phase" of said filters, not convolution related to room correction no?

Room correction is filtering like any other.

[vl]

13 hours ago, mcgillroy said:

Slightly OT but can you demonstrate this? Which CDs to compare and do you have any graphs too?

 

Here are two Sony Classical CDs that are representative of the Sony Classical CD sound.  As for Decca just pick any CD recording of the San Francisco Symphony Orchestra.

 

Wagner, Zubin Mehta, SK 45749

Hilary Hahn, Zinman, Beethoven violin concerto, SK 60584

 

The Sony Classical CDs sound like a microphone feed, but not the Decca or EMI ones.  The Sony CDs sound good on any CD player.  The Decca and EMI ones sound good only on the best digital playback systems.  I believe Sony chose a non brick wall AA filter for these recordings.  

[mansr]

1 hour ago, Le Concombre Masqué said:

what do you think of Cascaded integrator–comb filter ? doesn't moving average promoted by Ayre belong to that family?

A moving average filter has a rectangular impulse response. Its Fourier transform is the sinc function scaled by the width of the rectangle.

[vl]

7 hours ago, Shadders said:

Hi,

From the miniDSP sweb site :

REW uses IIR filters, while Dirac Live uses mixed-phase filtering - in effect, a combination of IIR and FIR filters. FIR filters are more powerful than IIR filters, but more expensive to implement. See the app note FIR vs IIR filtering for information on the differences between these two different types of digital filter.

(https://www.minidsp.com/applications/digital-room-correction/dirac-live-vs-rew)

 

I have not studied this, but one aspect that immediately occurs is that to implement room correction in regards to amplitude you will be splitting up the audio band into many smaller bands - so the gain of each band can be modified to achieve an overall flat response.

 

The band pass filters will always have out of band energy applied to them - and this will always be in the hearing frequencies (20Hz to 20kHz).

 

Therefore filters with ringing are a bad idea - you will need filters with no ringing.

 

Someone else may be able to confirm this.

 

Regards,

Shadders.

 

There are several issues at hand.  One is FIR vs IIR.  One is minimum phase vs linear phase.  Another one is filter ringing or not.

 

FIR takes more computation resources to implement and sounds significantly better than IIR.  I did a demo to my friends FIR and IIR crossover filters and they immediately heard the difference.  Most EQ devices and most miniDSP devices use the inferior IIR filters.  I was lucky to have found a miniDSP product that supports FIR at the 96k sampling rate.

 

Linear phase filters have constant group delay.  Sounds of all frequencies are delayed by a given amount of time.  Minimum phase filters have variable group delay.  Their latency is not fixed.  It is frequency dependent.  If we want to recover the information captured in a digital recording without amplitude and phase distortion, linear phase reconstruction filters should be used.

 

All brick wall filters ring, if they are excited at their Nyquist frequency.  If you do not want them to ring, simply put an effective AA filter in to limit the bandwidth to less than the Nyquist frequency.  These are all under our control.

[vl]

7 hours ago, Rt66indierock said:

 

I got to disagree with you as would Rob Watts and others. Preringing is necessary to reconstruct transits properly. 

 

I concur with this statement.

[mansr]

3 minutes ago, vl said:

FIR takes more computation resources to implement and sounds significantly better than IIR.

If the impulse response is the same, they sound the same.

[vl]

7 hours ago, Jud said:

 

As you noted, there is disagreement.  Some folks (including people not from the Meridian/MQA school) who actually do have some education and training in filter design use minimum phase filters because they think pre-ringing makes transients sound incorrect (since pre-ringing doesn't occur in nature).  Other folks agree with you and Rob Watts.  I hope to figure out a little more regarding what folks are disagreeing about by asking questions. 

 

Two points here.

 

1. Can a min phase filter do a perfect reconstruction job?  Look at this example.  I have a microphone feed with an upper bandwidth of 40 KHz.  I make a recording at 24/96.  Note that since the signal is band limited below 48 KHz, there will not be filter ringing.  Now I use a min phase filter to reconstruct the signal.  What I get will be distorted in phase.  If I use a lin phase filter for reconstruction, I shall not get distortion in amplitude and phase in the reconstructed signal.

 

2. If a recording engineer or producer is ill informed enough to use a brickwall filter to band limit a signal, he will capture the ringing of the filter.  In this case using min phase filters may make this brickwall limited signal sound less bad.  Personally if I come across such a recording (one should be able to tell ringing at 22.05 KHz on a spectral display) I shall just say goodbye to it.

[vl]

1 hour ago, mansr said:

If the impulse response is the same, they sound the same.

 

I agree.  Do FIR and IIR filters have the same impulse response?

[mansr]

2 minutes ago, vl said:

I agree.  Do FIR and IIR filters have the same impulse response?

Any response that decays to zero can be implemented as a FIR filter. It may be possible to achieve the same response more efficiently using an IIR structure. Categorically stating that FIR sounds better than IIR is a mistake.

[vl]

1 minute ago, mansr said:

Any response that decays to zero can be implemented as a FIR filter. It may be possible to achieve the same response more efficiently using an IIR structure. Categorically stating that FIR sounds better than IIR is a mistake.

 

Thanks for the explanation.  Let me clarify that the same 4th order 200 Hz LP and HP crossovers created by rePhase at a 96K sampling rate demonstrated that the FIR version sounded better - in transients and details.

[Fokus]

12 hours ago, Shadders said:

In terms of a rooms equalisation, where there are many peaks and troughs, and as per the text in their website - they use multiple filters.

 

But that does not mean that band-splitting is going on. It also does not mean that any of these filters need to be of very high steepness.

 

[miguelito]

On 8/30/2018 at 3:41 AM, crenca said:

Most of my listening (must be >90% now) is HP's, so no crossover (though I tend to use software crossfeed - a complicating factor?).  I am liking the sound of upper frequencies like cymbal, brass, etc. now that I am using  linear or "zero" phase most of the time, they seem to bring these out in a more accurate way, whereas there is a "smear", or softness when I use minimum phase.  I also think strings seem more subtly accurate as well. By upsampling everything to DXD and using the quality filters in HQPlayer, I am probably getting just about the best out of linear phase.  All this is andoctotal only...

Drive-by comment: Crossfeed is literally feeding some of the signal from the other channel, with a time lag if done properly, which I imagine the digital crossfeed would do. It is literally an overlay on the time domain so if anything I would think that it would make time smearing comparisons harder.

[Le Concombre Masqué]

7 hours ago, mansr said:

A moving average filter has a rectangular impulse response. Its Fourier transform is the sinc function scaled by the width of the rectangle.

 

Yesterday I switched to CIC as integrator in HQPlayer and liked it ; so you mean it can't sound better but for it's new and a change to my ears and that it has a different impulse response than IIR and FIR and sounds... square ?

[mansr]

4 hours ago, Le Concombre Masqué said:

Yesterday I switched to CIC as integrator in HQPlayer and liked it ; so you mean it can't sound better but for it's new and a change to my ears and that it has a different impulse response than IIR and FIR and sounds... square ?

I mean nothing of the kind.

[Le Concombre Masqué]

2 hours ago, mansr said:

I mean nothing of the kind.

and back to my initial question I simply wanted to know your opinion on CIC, SQ wise ; since you're debating FIR vs IIR, I was simply adding CIC to the discussion