Filter brewing for the Soekris R2R

[spzzzzkt]

Rather than clogging up the main vendor thread with posts and discussion about alternative filters, it seems more useful to corral all the information into one spot.


My intention is to put together a collection of basic filters starting with:

• Linear-phase anti-aliasing
• Minimum-phase anti-aliasing
• Linear-phase soft roll-off
• Minimum-phase soft roll-off.

At the moment I'm planning on working on 44.1khz versions of the filters as these influence the sound more than at higher sampling rates. I'll look at doing higher sample rates when I'm happy with the 44.1 files.

I'll post up response curves measured directly from the filter parameters I post up, and ask anyone else contributing filters to do the same, to make comparisons more straight forward.

If anyone wants to contribute analog measurements of the filters they are welcome to do so.

cheers
Paul

**********
Recommended Filters as at 1 June 2015
**********
NewNOS
"Correct" NOS filter implementation for the DAM1021
http://www.diyaudio.com/forums/digi...filter-brewing-soekris-r2r-6.html#post4245752

EQHQ v1
Brick wall filter, allowing aliasing to 23.5kHz.
Based on the stock 44.1 filter, but heavily reworked based on recent forum filter research.
All other filters redesigned and updated.
http://www.diyaudio.com/forums/digi...ilter-brewing-soekris-r2r-23.html#post4342582

EQHQ_apo and _quasiapo
Minor variations of the EQHQ v1 which limits imaging to 22.05kHz and 22.5kHz respectively.
These are more strictly correct than the v1, and have a slightly smoother character.
http://www.diyaudio.com/forums/digi...ilter-brewing-soekris-r2r-24.html#post4343625

**********

If people want to start then I now have the tools on our website:

http://www.soekris.dk/1021filt.txt -- ascii file with all filters
http://www.soekris.dk/1021filt.skr -- binary file for downloading onto a dam1021
http://www.soekris.dk/MKROM.EXE -- 32 bit dos utility to convert txt file to skr file, usage "mkrom 1021filt"

Please note that messing with IIR filters without knowing what you're doing can result in oscillations which will blow your speakers and/or ears.

The dam1021 FIR1 is operating at 352.8K/384K and each filter can have up to:

1016 tabs at 44.1K/48K input sample rate
508 tabs at 88.2K/96K input sample rate
252 tabs at 176.4K/192K input sample rate
124 tabs at 352.8K/384K input sample rate, but normally bypassed

FIR2 is operating at 2.822M/3.072M and can have up to 120 tabs, with input sample rates 352.8K/383K.

All upsampling is done by zero insertion, therefore gain need to be set to match oversampling rate.

There can be multiple of each filters in the 1021filt.txt filter file, but minimum one need to be there for each input sample rate. Currently just the first one for a given sample rate is used, in later firmware releases you will be able to choose between different filters.

Again, you need to know what you're doing if you want to do your own filters, it's not that simple and there aren't much error checking in the tools....

****************
Background Reading
****************

• Windowed-sinc filters (from The Scientist and Engineer's Guide to
  Digital Signal Processing - Steven W. Smith, Ph.D.: http://www.dspguide.com/CH16.PDF
• Anti-alias and anti-image filtering - Juilan Dunn: http://www.nanophon.com/audio/antialia.pdf
• Why do equalisers sound different? - Michael Gerzon: http://www.audiosignal.co.uk/Gerzon archive.html - download available in several formats.
• Ringing False: Digital Audio's Ubiquitous Filter - Keith Howard: Ringing False: Digital Audio's Ubiquitous Filter | Stereophile.com
• The Ayre MP Series (Digital filter white paper): https://www.ayre.com/pdf/Ayre_MP_White_Paper.pdf
• Theory of Upsampled Digital Audio - Doug Rife: http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf

****************
Test Tracks
****************
DPA Microphones - Miking A Grand Piano
Promotional CD illustrating techniques for miking a grand piano using their product range.
The was given away on request by DPA Microphones. This was released in 2007/8 and is long out of print.
Files are 44.1/16bit in .flac format and the zip archive is 270Mb.

As this is promotional material for DPA Microphones I believe there should be no issues sharing/spreading the word about DPA Microphones.

https://www.dropbox.com/s/223h1yuvlpx9s0u/DPA_MaGP_FLAC441.zip?dl=0

The SACD/CD contains 7 performances, with 8 different mic configurations.

1-8 Fallin' Leaves
9-16 Evening Song
17-24 Slowly Drifting
25-32 One for the New Ones
33-40 The Dat I Met Mr. Hayden
41-48 La Gazzellina
49-56 Minature XIV

The mic configurations are in the same order for each performance
1) 4041-SP, AB Stereo
2) 4006-TL, AB Stereo
3) 4052, In magnets on frame
4) 4011-TL, Bass/mid-treble positioned
5) 4021, ORTF Stereo
6) 4021, In magnets on frame
7) 4061, Hung from lid
8) 4061, In magnets on frame

The booklet that came with the disc can be download from the DPA site, but is mainly a discussion of miking techniques used.
http://www.dpamicrophones.com/en/Download/~/media/PDF/Download/grandpiano.pdf

 

[spzzzzkt]

Minimum-phase soft roll-off

To start the thread rolling, this filter is a second attempt at the Ayre Listen filter.
Unlike the MixPhase1 filter posted to the vendor thread it is pure minimum-phase.

I discovered the "trick" to minimising ringing is to stack filters in rePhase. By placing filters at approximately 2kHz intervals it's possible to create curve that replicates the desired behaviour using gentle filter slopes.

The filter impulse response is now very close to the 1 cycle of post ringing described in the Ayre MP white paper.

 

[rickmcinnis]

Hoping to get my DAC up and running so I can give your work a try.

I knew I could trust you when you said you use PUBLIC IMAGE/Album as one of your reference recordings.

It can tell one many useful things plus it is a great record.

Are you able to enjoy the latest one?

 

[spzzzzkt]

Hoping to get my DAC up and running so I can give your work a try.

I knew I could trust you when you said you use PUBLIC IMAGE/Album as one of your reference recordings.

It can tell one many useful things plus it is a great record.

Are you able to enjoy the latest one?

I haven't had a chance to listen to Album with the new filter. I'll have a better chance to listen tomorrow, but at the moment I'm playing Bill Laswell's "Means of Deliverance" (solo acoustic bass), and my initial impression is this is a clear improvement over the MixPhase1 filter.

cheers
Paul

 

[picovolt]

Paul--thank you for doing this (both the thread organization and the filter experiments). I'm still piecing together parts of my board, but am really looking forward to the filters that will be available. I've been using Audircana+ for awhile, and the filter adjustments in that have almost put my turntable up in the attic.

 

[rickmcinnis]

I meant the latest PIL record though the question very well could have been a double entendre ...

 

[spzzzzkt]

I meant the latest PIL record though the question very well could have been a double entendre ...

I was tossing up which interpretation to answer....

No, I haven't.
I have to admit I was a Bill Laswell completist for a while, which is one of the main reasons I own Album.
That said, it's an incredible lineup playing on that recording.

 

[rickmcinnis]

Even if one is not a PIL fan that record is easy to like.

I really like them, especially the second record.

One great thing about Lydon is he is always changing so when ALBUM was a departure one went along with it.

Thanks for your explorations.

Went to an electronics store and bought the wrong gender RS232 cable so tomorrow I get to hear the DAC with correct firmware and your filter.

 

[spzzzzkt]

"NOS" filter

Not too much you can say about this really.

All sampling rates have a single filter tap set to 1.000 which effectively means they are a bypass.

44.1 and 48khz = x64 up-sampling
88.2 and 96kHz = x48 up-sampling
176.4 and 192 = x16 up-sampling

I haven't adjusted the gain, so you'll need to turn it up a bit.

No graphs, as it's dead flat response, and the IR is a single value.

All that remains is Søren's final FIR on the x8 upsampling, which I don't plan to touch at this point as it filters out content above 220Khz.

 

[spzzzzkt]

Linear-phase Slow Roll-off

This is a similar filter to the Ayre Linear-Phase "Listen" setting.

As with the minimum phase filter, the main aim was to get as little ringing as possible.
The stacked filter technique seems to help in achieving this goal.

 

[TNT]

The LPSRO441 above has quite slow rolloff. Its only some 10 dB down at 22 kHz. Is this really enough? Energy that is present above 22,05 (1/2 Fs) will reflect back into the 0-22.05 range.

So the resulting alias from your filter at 29 kHz will turn up as 29-22=7khz. At 29 kHz you are down only -15 dB. This is why "everyone" is aiming for -100 at 22,05 khz.

During listening session yesterday over a pair of JBL Array 1500 and Hypex NC400 one can hear a noise trail around higher notes quite clearly. It creates a very "clear" sound but also it suffer from digitalis and really isn't clean.

Are you sure you mimick the Ayre at 22,05 khz?

//

 

[spzzzzkt]

Yes. The Ayre filter is -6dB at 22.05khz. -100db at 44.1khz.

However, re-reading the WP this seems to be an older filter design which is superseded by the MP slow roll-off.

The first approach to solving the problems of a conventional digital filter was to use a filter with less ringing, often known as a “slow roll-off” filter. This type of filter was used in the “Listen” position of the original Ayre disc players. By reducing the “sharpness” of the “knee” in the filter’s frequency response, the filter’s transient response is vastly improved. Now there is only about one cycle of pre- and post-ringing.

The penalty (remember, there is no such thing as a free lunch—only intelligent tradeoffs) is that there is more “leakage” (aliasing) of high frequencies above 22,050 Hz back in to the audio band. Still, this only affects very high frequencies and the levels are low enough not to cause audible problems.
This “slow roll-off” filter reduces the time smear by a factor of ~20x compared to conventional digital filters. The net result is a much more musically natural sound, as the ear-brain is very sensitive to time-related distortions. This filter provides an outstanding compromise between frequency response and transient response, and for ten years was the mainstay of Ayre’s digital audio filters.

cheers
Paul

 

[TNT]

Åhå. And this is all their filtering. There are no additional analogue ones?

//

 

[aive]

Does anyone have any idea wtf Mike Moffat of Schiit Audio is talking about wrt the bit-perfect audio filter in their upcoming DAC? Just wondering if anyone can work out the algorithm for the Soekris R2R and if it's as magical as they suggest.

An SOF (Schiit only feature) – The Schiit Footlong Mega Burrito Supersauce Digital Filter:

It is a digital filter/sample rate converter designed to convert all audio to 352.8 or 396KHz sample rates so that it may drive our DACs. You get it from us; it is our filter. It keeps all original samples; those samples contain rudimentary frequency and phase information which can be optimized not only in the time domain but in the frequency domain. We do precisely this in the Yggy with said filter; this is the reason that on good recordings through Yggy you can hear the hall, its dimensions, and the exact position of anyone coughing or farting in the room, the motions of guitars being hoisted in preparation of being played, sheet music pages being turned, etc. etc. This comes from our mega burrito filter. A friend of mine, Jonathan Horwich, sells analog master tapes in ½ track form – at least 15 IPS, and 30 (I believe) as well. On those analog masters, you can also hear the entire environment before the music starts – what is amazing there is that even if on accounts for hearing “down into” the analog noise, the S/N indicates a 14 bit performance at best for those tapes. 14 bit or not – those tapes, totally scratch my itch. If you want that, we got that and more in the Yggy.

Thoughts on a bunch of DACs (and why I hate chocolate ice cream) - Page 121

 

[TNT]

Some god marketing there. He claims that interpolation is a dead exact exercise - good luck with that. On close(r) recordings without deadening mastering you will hear people change page in then notes. Nothing special about that.

//

 

[zfe]

The LPSRO441 above has quite slow rolloff. Its only some 10 dB down at 22 kHz. Is this really enough? Energy that is present above 22,05 (1/2 Fs) will reflect back into the 0-22.05 range.

So the resulting alias from your filter at 29 kHz will turn up as 29-22=7khz. At 29 kHz you are down only -15 dB. This is why "everyone" is aiming for -100 at 22,05 khz.

The 29 kHz will turn up as 44-29=15kHz (first we are in the lower side band which is mirrored frequency wise).
But still I also feel that the rolloff is too slow.

Åhå. And this is all their filtering. There are no additional analogue ones?

//

It would be a strange design concept to combine slow digital filter with a fast analogue one. Ususally the arument is you can do the fast ones more precise digital.

 

[TNT]

Looking at Ayre

Meaure:
http://www.stereophile.com/images/archivesart/1009ayre.fig12.jpg

Listen:
http://www.stereophile.com/images/archivesart/1009ayre.fig13.jpg

The difference is more alaising product but not at an alarming leve but look at the main IM product at 1khz. Now, today we don't build analogue circuits that produce -55 dB IM so I reckon this is a digital artifact coming out of alaising distorsion via the digital domain.

These things make the sound appear clear but is tiring in the long end.

/

 

[spzzzzkt]

Åhå. And this is all their filtering. There are no additional analogue ones?

//

I can find no description of the Ayre using additional analog filtering. I can't find Charles Hansen discussing anything but digital filtering, but they do use a discrete output stage so there may be something incorporated into that.

I'll let you find evidence that there is additional filtering as you've made the claim that there might be...

 

[danny_66]

Thanks Paul for creating these filters, those customizable filters makes the Soekris R2R much more versatile.

When searching for information about filters I stumbled upon Auralic Vega DAC, it has different filters you can select.
Most of the users prefer filter n°4 that is also a minimum phase.
Here some links for additional inspiration

Stereophile measurements of different filters

Hometheaterhifi measurements of different filters

AURALiC Flexible Filter white paper

subjective observations of the different filters

It's time to put my Soekris R2R DAC together

 

[TNT]

A few contenders...?

A friend of mine has created a few different filters. I will later present their characteristics. It would be interesting to hear what you think about them.

Here comes the first.

//