'FeralA' decoder -- free-to-use

[John Dyson]

About the planned SW release tonight...

Even though the demos sound right to me, I have a strong belief that my hearing is now especially unreliable.  Instead of doing the release and potentially wasting someones time, I'll make the demos avaialble again and let people waste their time in doing some of the comparisons.🙂

There is no-one available who can do the needed verification, so I'll wait until morning (luckily I am back to 4Hrs sleep), and check the quality for myself.

 

I CAN hear the dynamics, and so far they match pretty well, except the decoded version generally has stronger dynamics, as expected.

You cannot believe the frustration right now -- I'll bet you I have loss at 6kHz now.  I couldn't tell the difference if I had to.

 

The bass is essentially same as the FA, except the decoded version is more tight as expected.  FA IS compressed bass, and additionally it adds higher frequency distortion to tell you to look for bass.    On decoded material, it requires that one hears true bass that is just that magical thing that vibrates without the higher frequency envelope of distortion.

 

I'll quickly post when the demos are ready -- otherwise, signing off for the day -- my typing is really bad.

 

 

[John Dyson]

The V2.2.5J-0 demos are ready, but still a little disappointing.

(This will be blather into exhuastion, but really wanted to give a complete, honest, transparent status report.)

 

The MF deadness is fixed entirely.  Snippets same place -- still imperfect, but less imperfect than previous:

https://www.dropbox.com/sh/tepjnd01xawzscv/AAB08KiAo8IRtYiUXSHRwLMla?dl=0

 

However, now there is another kind of deadness, and I should have noticed it during my tests earlier, so

unless you are really curious about the step by step progress, you might or might not want to download some of the V2.2.5J  examples.

They are not a waste of time, just don't expect the practically total perfection that I personally expected.

 

* These comments about HF comp on this post should be taken with a grain of salt, but I am assuming that my

perception is working 'well enough'.  If

I break the code because of fixing a bug that doesn't exist, we can quickly revert..   3 machines and at least 6disks have every version

ever produced, every in-between version, and most definitely the last version that was pretty close.  There is a full, online

history of well over 1000, probably 2000-3000 versions, most ready to run, and go back to approx 2014 or 2015.

@KSTRis gonna make me use code control and management tools 🙂,  (actually he  rightfully emphasized and

reminded me) but for now, beause of extreme cheapness of HDD and SDD,

I have about 6 copies of the code, spread over multiple machines and multiple disks along with

temporarily mounted copies.  (This backup is done autmoatically when a da-Linux version is created.)

I do agree with @KSTRsentiments, and fully agree with the need to do what he suggests.  It is on the list for

implementing very soon.

 

The technical correction for decoding results seems again in the EQ for the input.  The  audio problem  can seem slight or more

than slight from recording to to recording.  This is probably an EQ driven dynamics issue, but  instead the MF, now in the HF.  The

change will almost be freq response balance neutral, giving a just slightly brigher sound in general.  This EQ issue is not easy to detect

based on freq balance alone.   This is very similar to dynamics issues often causing profound frequency response balance

problems.  Neither frequency balance nor dynamics can be separated into totally orthogonal/independent entities.

This bounce back and forth mechanism is mostly NOT needed if there were specifcations or schematic.  Without

the decoder output somewhat closely  matching the original from which the FA was made, we would  not have detected

these errors.  These kind of changes are 'tells' for the improvement in frequency response balance accuracy, even

though finding these response balance problems can be irritating.

 

The current test/results behavior (are 'tells') are what I have been looking for them for a VERY VERY long and painful time!!!

I am truely happy about the current status right now, even though not perfect.

 

========================
Lets bounce back and forth over and over again. 🙂  Each time, we are getting closer and closer!!!

The general sound IS indeed getting closer and closer, as hoped some day.  The various noise and fuzz

enhancement of FA are greatly reduced, almost back to normal, pure recording. 

 

The current decoding flaw showing weak dynamics in the highs has a very different feel than weak dynamics in the  midrange.

The midrange deadness is much more irritating and noticeable. The weak dynamics on the HF are a different kind of deadness,

but not quite as irritating.  HF weakness makes you want to turn up the treble, but that will never fix it very well.  

 

In the recent MF case, the solution was obvious to me

that the input had to have the MF dip, while the output had to have the compensatory reverse-dip (not really a peak.)

 

For HF correction, will be almost opposite to the MF correction. It appears that this new issue will be better

corrected by a rising HF response  that is then compensated on the output side  by a decreasing response.

So, most likely, but I can be wrong, on input, there should be a rising response at 3kHz +6dB to add to the decreasing response

at 9kHz that is already implemented.  On output, like other pre-emphasis, there will be a complementary de-emphasis.

 

We previously had  this rising 3kHz EQ, because it seemed correct.  I removed it because of  HF 'over-enhancement.'

With the new MF scheme just added, my original intuition about the need for HF compensation will probably be vindicated.

However this vidicated scheme was forced by the MF EQ, not needed when the MF is not correct.  Isn't this stuff

wierd?  No matter what,  ALL OF THESE EQS are EVEN VALUES.   Some of these values are so critical that the wrong choice even by one

step, in the standard scheme of things, might damage your hearing, or might become so attenuated that it is difficult to listen to.

 

I am going away until tomorrow unless there will be an announcement about 'beautiful music' :-).  Tomorrow, I am hoping

to have prepared a nice toy for everyone during the weekend, but we never know what will really happen.  I thought that this

thing was working 1.5yrs ago.  Cannot trust my judgement on this program being correct!!!

 

 

[fas42]

Okay, John, it's been a long time since I tried listening to your standard set of clips - so, revisited Mamma Mia, by ABBA - to see where we're at ... 🙂.

 

Right, much better than from what I heard the last time I checked what the decoder was doing - the last time, many, many versions ago, the damage done was far too great - and I said this, at the time. This round, the coherence of the song is basically sound, but, it has been stripped of energy - the richness and harmonic density of the intro now sounds somewhat awkward, as if the producer didn't know how to make it more engaging - what has been done, is to suck the energy out of the backing instrumentals, which means that the vocals are more strongly highlighted; which are fine as presented here. But you can still hear that the tonality of the instruments is affected negatively - particularly clear is that the cymbal work on the drums is strongly muffled; as if a cloth had placed over each such instrument for doing the recording.

 

 

[John Dyson]

(I have read only a few messages today, no insult intended -- trying hard to make a good toy for everyone.)

 

There is absolutely no reason to waste your time on V2.2.5J.  I think that I just heard Angels sing on the K test version, and it was an ABBA recording (no -- not any mood or perception altering  drugs :-)).

 

I can never proclaim perfection -- I need the users to tell me that when/if it happens.  Also, remember my totally garbage hearing at times --

my  hearing currently is somewhere between worst and best.  So with a reasonable scepticism,  there is some possibility that the decoder is working

quite reasonably well.  I promise that it is NOT the final version, but it has moved forward.

 

You know how I emotionally exaggerate, so take my proclamations like this, with a grain of salt, but after the HF EQ had been re-instated after the new MF EQ was added,

it really fixed a lot of the negatives that worried even me.  (I am talking about both dynamics and EQ issues.)

 

I am not going to hurry this tonight, but hope to make viable demos by tomorrow morning my time (about +14Hrs.)  I don't sleep much beyond 4Hrs, so this goal might be possible.

If some serious focus on double checking, quality control, etc can be sustained -- I am hoping that a useful release might be produced by tomorrow afternoon my time, trying to sneak it in before our EU and UK friends go to bed.  I keep hoping that things continue to work well, and this time

we might have something.  Please don't find any serious problems -- however, if you do -- I really NEED you to help me understand as well as possible.

 

(Removed a lot of rambling, you do get the message, I am sure. Finished trying to type messages for the night.  I must focus, hopefully benefiting all of us, if not tomorrow, soon after.)

 

 

[John Dyson]

The version just ran off might be acceptable to even some picky people.  (I am also picky, but maybe in an eccentric way.)   However, I thought that I did something smart and opened up the very low bass -- trying to give a better sensory impact -- getting a little closer to the agc boosted bass on FA.    For some reaon, I didn't consider that  controlling 1st order EQ, when trying to change sound, can be enormously tricky.   Instead of just get a better 'deep, magical' bass, there was also a slight increase in that muddy sound.   The error isn't technically very large, but really does sound like an error.

 

Instead of wasting anyones' time with yet another upload/download, I'll be delaying the decoding demos by one to two hours.  The current results do sound very, very similar in many ways with the FA version (sans hiss, etc), but I am bass adverse, and ESPECIALLY adverse to excess bass.   I am trying very hard to avoid uploading excessively expermental examples this time, and this problem is almost trivial, but tricky at the same time, to fix.   This corrective change will likely be at 9.375Hz with 1st order filters or 18.75Hz with 2nd order flters with Q between 0.8409 and 1.414.   Changes of EQ at those frequencies have very noticeable control over certain kinds of 'muddy' bass.  In my perception, this muddy bass does obscure vocals a little, and I believe that mimicking the sense or intention of the vocal without the 'telephone' effects is paramount.

 

So, I'd expect the upload to be start about +2Hrs, maybe as long as +3Hrs from now. I really don't expect a delay for the demos go to beyond +2Hrs as that would indicate a problem, but giving the 3Hr window will allow me to be distracted on personal matters.

 

Also I want to do another once-over on the decoder before uploading da-avx/da-avx512/etc, and unfortunately, right nowthere are limitations to changes that I make when I am running demos.   So, the binary wont be avaliable until as early as +7Hrs from now, I am trying sneek it before most people in UK, EU, etc go to bed.

 

=====================================

The rest of this is mostly optional details -- don't bother to read if you are not interested.

 

 

Of course, even updating the version number requires creating a new binary -- so double checking and desirable  change/test must be delayed until building the demo examples is complete.   Because of a lack of foresight when creating my current set-up, just reviewing the code is okay, but building binaries can sometimes interfere with the decoding operations.   As a result of a non decision when setting up build environment I didnt' create a 'safe space' for my test binaries.   Probably unexpectedly, when doing the big decodes, I use the exact same binary as I build when testing (acutally, for the demos, the da-avx512 binary is used).   In a more well thought out situation, I would have moved the binary to a 'frozen' location, but I didn't.

 

Right now, I am listening to the 'test-distribution build' results and TO ME, they sound good, but there is a problem from slightly (and I truly mean SLIGHTLY) muddy bass that interferes with the clairity of some vocals.  With the limitations of my perception, we are fairly far into 'at low volumes, the result often sound very similar to FA'.   At my normal listening levels, the difference is very obvious -- so *by my perception* we are getting close to where we should be.*   Eventually, if we all agree I'll add a control switch (maybe two) to do any needed fine adjustments, and when these results are reported, I'll make the associated final changes.

 

It is getting tricky to make changes now because of my 'rules' about using a standard set of EQ methods and parameters.   The steps are sometimes being more coarse than 'good sound' can allow.   An example of this, is that I am making the 'step changes' at 18 and 9 Hz to control the muddy bass.   There are no reasonable steps left at 37.5, 50, 75Hz.   From what can determine any properly coordinated changes at 37.5,50 and 75Hz would sound worse than the current.   My control at 18.75Hz 2nd order is pretty much limited ot Q value, because any changes to the shelf gain/attenuation parameter pushes us to outside desirable sound characteristics.   Also, another thing about the 18.75Hz control scheme, but listening for the few LF tells that I do have, a series of three of the same 18.75Hz 2nd order -3dB EQ, the only variable now is Q.   Right now, I am using 1.414 (the highest usable value)-- trying to make the attenuation curve tighter against 20Hz, but instead we ended up with too much bass.  (For the EE experts, the Q=1.414 doesn't result in much of an actual gain peak because there is already -3dB at 20Hz, so the actual peaking effect is not quite as evil as one might intuitively think. )   The important correction isn't actually much about the gain at 20Hz, but more like the effects at 30-50Hz...   By making changes at 18Hz, there might be a slight but also important change at 30Hz.   Like I have written over and over again, this EQ is really tricky.   It is MUCH less common sense than one might initially think!!!  The EQ really has to be carefully considered, as much about how to get the precision as the specific required value itself.

 

* After all of these recent troubles, I trying to avoid much reliance my 'golden ears' (humor intended) about frequency response balance.

 

Will be busy testing the decodes as they come off the 'assembly line' so that I catch any new or previously heard problems as quickly as possible.

Thanks for your patience, both today and for the last couple of years.  (For some people, it has been longer than that.)

 

I expect, at the very worst +3Hrs from this posting time.  (not considering medical emergencies, highly highly unlikely.)

 

 

 

 

 

[John Dyson]

 

* At the last minute, before producing the snippets from the raw files -- I listened carefully, and heard a problem 'tell' that will only happen if there is not enough HF EQ.

I decided to move from 18kHz/-3dB to 18kHz/-6dB, and that tell went away.  I did not run into that problem until listening to a specific recording.  In this situation,

and knowing what I know now, the only possible *simple*/*likely* candidates are 18kHz/-6dB or 12kHz/-3dB, but the 12kHz/-3dB had some problems that apparenlty 18kHz/-6dB

does not.    This is a very, very frustrating situation for me right now.  IF there is still a problem in the demos -- I ask that you tell me right away about the HF, PLEASE.  I will know almost immediately what will need to be done, once I know what the signal sounds like.   For now, my 'meter' has quit working. (humor with sadness intended.)

 

I really paniced this morning, because I am not sure that the HF that I heard last night was actually good.  Don't get me wrong -- the sound IS good, just that hearing HF is a problem

right now.  I cannot trust the HF characteristics.   Once there is a modicum of feedback, the answer will be in more likely correct decoder today.  The feedback this time really

needs to describe the HF sound -- I cannot do HF by myself right now.

 

* I will continue to try to detect which choice is correct -- this is a really irritating situaiton, because the decoder, not being perfect, is so close to being GOOD.

 

After reviewing some others, trying as hard as I can, that EQ could still be sub-optimal, but the other nearby standard and also plausible EQ

given the good enginerring choices was 12kHz/-3dB, but that had some problems that I COULD hear.  I really don't mean to imply that there is a probable

problem -- it is just that I cannot make a good choice.

  I

Now, I am doing the re-decode in a lower quality mode, so it goes really quickly on my machine, but is still one of the  'high quality' modes.   Just that it is a less high quality

mode than I normally demo or use myself. (It is about 3-4X faster than realtime, instead of 1.2X faster.)   This --fz mode still does anti-MD and anti-IMD, but

is less heroic.   'Take a Chance on Me' is the only recording in this set where the distinction between '--fz' and '--fz=max' makes a very noticeable difference for me.

 

My HF hearing is GONE, and all I can do is to listen for tells, and try to compare with the FA as carefully as possible.   When the demos are made ready (ASAP), there

will be an attached 'full reportt' about the last 24Hrs.  The report/background info about what was done with the decoder and reasons will be attached.  I already create enough noise.

 

[John Dyson]

I really regret the delays, but the snippets are now ready.  I had to use a slightly lower-than-usual decoding mode, but is still 10X a DolbyA in quality, if we had used DolbyA HW.

 

Full RAW and DECODED:

https://www.dropbox.com/sh/tepjnd01xawzscv/AAB08KiAo8IRtYiUXSHRwLMla?dl=0

DECODED only:

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

There are two versions of the snippets.   One directory has both versions.  The other is decoded-only.   I have a scheme where the interleaved versions are played automatically, but also listening to decoded-only can be very instructive.   The Decoded-only versions are exactly the same as in the full package, I simply separated them to help where I can.

 

I have several dB of control on the low end and a few dB on the high end.  Most of the control left over is about placement, not as much when it comes to 'dBs'.

This matter is mentioned in the attachment,that sometimes changing/correcting EQ is not intuitive, esp 1st order EQ.

 

When comparing, please understand that there are innate differences between decoded and FA.  In fact, one of the possible devices that the FA decoder might be correcting is called the 'BigBottom' device.   Think about the FA bass sound, and the resulting decoded sound.  I really doubt that it is a 'BigBottom' because of the complexity and very precision behavior needed in the decoder...   The different bass DOES come from compressing the bass, and not all of it can be an error in the decoder.   Giving more bass in the decoder does produce other side effects, and it has no advantage of the compressed recording.   There is really a good reason why the encoded/FA version might be desirable in certain cases, but going very far in changing the decoder beyond correct cannot happen.   IF we can figure out what is going on, I am very happy to add more bass.  In fact, while listening, I know how to add more bass without making it muddy RIGHT NOW, but will it be correct?  YOU TELL ME!!!

 

IF WE DO NEED THE STRONGER BASS, I'll add what I can in the upcoming release.

 

I am VERY bass adverse -- so I need help in making a decision that everyone likes.  As you can probably tell, the response of the decoder 'goes all the way down', but there is some control to add bass at 80Hz without tubbiness (I hope.)

 

Sometimes it is best to communicate privately, but ANY criticism is okay publically.  It is just that there is sometimes more freedom when communicating privately.

 

Also, in the attachment, there is a spiel about what I have been doing on this stuff, and ends up being a reason why I need help with the EQ today...   Also, it contains an expose about how I must operate when I don't have any kind of specification at all.   Bottom line, it is almost like mind reading the design engineer.  (Not really, but it almost is metaphysical.)

 

A little help in deciding about the final HF/LF EQ would be so very nice...

 

(000) Info about V2.2.5Q demos & V2.2.6A release.odt

[John Dyson]

On 4/15/2021 at 2:28 AM, PeterSt said:

@fas42 Frank, I could also say: If one likes the RAW better than the DEC then something is really amiss because that should not be so at all. As you have seen, it is my conclusion too. However, in comparison with the originals, the RAW versions **** (I am not allowed to use the word here because John goes bananas of that word and requires further explanation what I G-D provided).

 

I have been begging for input...  'It sucks'  or a technical babble-talk equivalent helps no-one though.

 

[John Dyson]

On 4/15/2021 at 4:25 AM, KSTR said:

The original master is from 1952, no DolbyA ;-)

On the CD releases, they may or may not have used DolbyA to pimp it, whatever n-th generation copy of the master they had as source. The source could as well have been a professional vinyl rip with all the bells and whistles.

 

What we know:

- John's raw version has been resampled to 48kHz (why?)

- decoded version(s) (which would have been 88.2kHz) have been resampled as well to 48kHz (why?)

- decoded version(s) have the JD house EQ as ususal, making un-skewed comparisons very hard (IHMO)

 

To check out which CD version John used (unless he's telling us) would be to use DeltaWave to check against available CD versions. DW takes care of the resampling back to 44.1, and any larger differences in the mastering would quickly pop up (setting filters at 20kHz to keep the resampling stuff out of the picture would certainly be needed).

 

The only reason why it is resmampled is because it is exactly the same processing that the 88.2k/FP version goes through to produce the snippet.   The only difference is the necessary rate conversion.   In these specific situations, because of the ease of scripting, I am using SoX with the rate in -v mode.   The needed metadata savling/restore because of SoX heavy hand is done by metaflac and/or sndfile-metadata-get and sndfile-metadata-set (forget the exact names.)  The DHNRDS is very metadata friendly, and even creates/manages the BEXT data by default.

* almost exactly same processing between CD->snippet as decoder->snippet

In reality, the playing field is very even.

 

It is NOT an intentional 'house' EQ, but the decoder NEEDS EQ because DolbyA units don't work well in this application without it.  After listening, I would suspect that almost anyone would believe that the recent corrections are better.   The decoder was missing an entire pre-emph/de-emph step that could not be inutited by frequency response alone.  When adding that pre/demph in, then a previously removed pre/de-emph was readded, and fits in as I had previously intuited.   This new MF EQ  is related to the necessary truly very precise EQ needed between layers.   If there isn't a 9dB 1-3kHz dip between each layer, the levels go totally bonkers. Likewise, without the 500Hz,-3dB and 75Hz,-3dB Eq between layers, the bass goes bonkers.   These freq/dB numbers need to be precise.

 

One might reasonably proclaim (I truly intend to mean -- REASONABLY) that it is just DolbyA stuff being corrected.  Well, how can that be?, there is very little in a DolbyA that is an accurate 'even dB' number.   The only one that I know of, that is an 'even dB' value is the 0dB at high signal levels.   The EQ correcting something from the design engineer.  DolbyA units are siimply not 'even dB' devices.  As far as I can tell -- and I have done the reverse engineering, the 80Hz, 3kHz, 9kHz are not dead on.  The Q values are not even numbers, the gains/thresholds on each band are not matched to even numbers, and the maximum dB loss on each band are not even numbers.   It does appear that DolbyA was 'designed' by building hardware by someone with genius engineering intuition.   Some of the portions might have been designed (attack/release circuitry/algorithsm) but the DolbyA was obviously not designed in the era with ubiquitous computing being commonly available.

 

When the EQ is not correct, that is mostly  about problems coming up with the last phases of EQ.   In fact, as noted above, in the last few days, by listening to a very nearly correct decoder, I could natively hear that pre-emph/de-emph was needed because of the problems with the dynamics.   Likewise, once the new MF EQ was added, I heard another set of errors in signal dynamics, and that additional correction was something previously designed in by me.  I had to 'mind-read' an old, likely dead engineer (this figuratively accurate) to decide what kind of additional, non-DolbyA specific pre/de-emphasis was needed.   Very interestingly, this new HF pre/de-emphasis is almost exactly the same Hz and dB as the original that I removed.  The HF EQ that I just re-inserted was originally correct, and all of original use of the EQ came from my understanding of what a combination of DolbyA units would need in this situation.   One also must add in with a guess about what a responsible engineer would design.   There really is/was a lot of educated guessing in the project, and it was hard-earned, hard fraught, and totally exhausting to do.

 

Once I could hear the dynamics from an otherwise relatively accurate decoder, I knew EXACTLY the correct MF eq, down to the circuit configuration and circuit values.  (Whoops, some of the EE in me just popped out!!!)

 

Understanding what/how the decoder *should works* is totally solved.   Guessing the optional choices in the design can partially be  intuited (e.g. the bounds), but truly needed to be measured based on source material.  We just fixed a major shortcoming that could not have otherwise been solved.  (A schematic or spec would have really helped.)* A few hurt feelings would not have happened without some very frustrated people focusing for so long to solve such a complex problem...

My brain really does sometimes hurt.. :-)

 

* with a spec or a schematic, even just an accurate circuit diagram without parts values, this project would result in functioning software in about 2 months.  This is all about painful reverse engineering, guessing, and some seances. :-)  I hadn't made any deals with the devil yet -- never really tempted, but I thought about it :-).

 

 

 

 

 

 

 

 

[John Dyson]

Here is a comment about the demos and the upcoming release.  I am in the midst of deciding whether or not it is correct to add the approx 1.5dB@80Hz Q=1.0 EQ for the bass.

When listening to the snippets, please tell me if you think that additional bass would make the signal more correct (in your opinion).   I am adding the EQ no matter what, but will either make the EQ enabled by default or disabled  by default.   I want to make the decoder as fully functional as possible as soon as possible.

* BTW -- this specific EQ characterstic is architecturally valid, all depending on what the original designer had chosen.

 

After some work this afternoon, the EQ will be inserted, my only decision is whether or not to make using that EQ as default.  It might even be wrong, but without additional info, I'll make my best guess.  I just thought about the EQ as I was uploading the examples a few minutes ago.  I don't like a lot of bass, so I wasn't as focused on it as I should have been.

 

Once I seriously thought about people other than myself,  I immediate knew what was needed make the decoder more 'bass friendly'.   Perhaps my biggest resistance to even think about the needed EQ is that this is one of the few places where 2nd order might be valid.  I am really mentally locked-in to using 1st order EQ...   using 1st order effectively is more of a challenge anyway, and I prefer challenges :-).

 

 

[John Dyson]

I am resceding the demos.  I should NOT have touched the 12k/-3dB EQ.   My hearing just came back...

Can you imagine how frustrated I get?

Can you even imagine a LITTLE about how irritated I am right now???

 

In a few hours, I'll do a decoder release (with everything that I talked about), and the demos will be available in similar timeframe.

 

I'll leave one or two examples on the site, I reallly don't want people to waste their time.

My hearing f-worded me again.   You can hear that the EQ is wrong (if you can hear at all.)  It sure sounds like I needed the original 12k/-3dB.

Sorry about this -- I am trying so very hard.

 

[KSTR]

1 hour ago, John Dyson said:

* with a spec or a schematic, even just an accurate circuit diagram without parts values, this project would result in functioning software in about 2 months.

https://audio-circuit.dk/wp-content/uploads/simple-file-list/d-other/Dolby-361-sm.pdf

Unreadable part values, though. But good general information.

EDIT: A way better readable schematic is located close to the end of the manual.

[KSTR]

With this manual, the path is open to do a Spice sim with LTspice. LTspice can use .WAV files for time-domain input and output, so actual music files could be rendered, and then decoded and see how close one gets to the original.

Only problem is those JFET models and trimming their parameters so they match what Dolby had selected them for, though some educated guesses sure could be arrived at.

Even better scan here: https://www.richardhess.com/manuals/Dolby/Dolby CAT 22 Schematic scan 01.pdf

[John Dyson]

1 hour ago, KSTR said:

https://audio-circuit.dk/wp-content/uploads/simple-file-list/d-other/Dolby-361-sm.pdf

Unreadable part values, though. But good general information.

EDIT: A way better readable schematic is located close to the end of the manual.

You are correct, about the DolbyA.   I did that a long time ago, and really did stupid things until I really studied the schematic and carefully studied and understood the circuit that *looks* like a normal full wave detector with some reverse bias protection for the transistor.   Really looking very carefully, it definitely is not.   An initial hint for the 'unitiated' is the carefully biased diodes.   That is one hell of a tricky design for what it actually does.  Even for what the circuit does, which is very complex in its own right, it is very deceptively simple circuit.

 

The big stumbling block, for emulating it, if one does generally understand the circuit, is that critical and important parts are selected.   There is literally NO WAY that an emulation could be done without analysing a unit in hand.   I learned a lot of 'lessons' about emulation and reverse engineering in general when doing the DolbyA and the FA project.   FA and DA are actually two totally separate projects, and the DA decoder is not dependent on anything about FA,but FA is VERY dependent on DA.   Even before I knew the low level details for FA, where my expertise is still very suspect, I did see a lot of DA characteristics in the FA signal, and after learning probably everything about DA -- and reviewed the conjecture again, I ended up being 100% positive without  any equivocation that there is definitely DA technology actively used to produce the FA signal.   From the information being talked about in this specific note, I could not say that a 'DolbyA' was used, just that something with the same characteristics of a DolbyA is being used, even if it is just a subcircuit.

 

Once I quit trying to make a superior expander doing the DA  project,  realizing that the DolbyA encode/decode is very very hand-in glove,and has NO slop-factors like DBX does - I decided to design and implement a truly superior Decoder, while carefully emulating the actual HW units where it is critically important.   This might sound a little wierd, but I have some ethical resistance against doing an encoder, so I didnt' bother.   An encoder would actually be  fairly siimple given the already developed technology.

 

DolbyA encoding is very good quality.   DolbyA HW decoding is actually sloppy.   This is because it is implemented by the encoder being in a feedback loop to reverse the processing.   It seems to be okay, and really pretty cool to perfectly reverse the encoding.   Intuition will leads a person astray, because if you put your engineer hat on and really think about it -- lets ask the question?   What is a major nemesis for feedback circuits to work, even  worse than phase delay?   Well, the encoder design has enough scrambled phase delay that you really have to think of it as a time delay.  The encoder being in the feedback loop produces a result that is sometimes of suspect quality from a purist standpoint.  This delay is even worse than one might intuit when it comes to signal quality damage.

 

After doing all of the math to emulate every little nuance of a feedback loop based decoder, I have selectively transferred whatever delays/scale factors into my feedforward design, but where it impact the signal negatively, I didn't transfer that behavoural aspect.

 

Even without all of the fancy processing that has been added, other than a few frequency response bumps, even dynamically, the DHNRDS comes very close to a real DolbyA, but without it's decoding flaws.   The additional fancy processing helps on normal material, but with the fancy DA addon processing,  it probably makes FA decoding possible with reasonably quality.

 

I understand that there are LOTS of DolbyA schematics out there, there is even a 301A, which is a totally different animal, and I probably should have emulated that one, because it is all diodes instead of using very widely speced JFETS.   If emulating diodes, you can get both the repeatable exponential curve for gain control, yet not emulate the nonlinearity for the signal.  When I started, the cat-22 style DolbyA just seemed like a simpler circuit -- if I was to do it agian, I would have looked at the 301A more carefully.  (I might have remember the product code incorrectly -- it has been a few years since I referred to much DolbyA stuff.)

 

Also, in my secret storage area, I have a full, complete SR schematic -- those are very hard to find.   It is an ugly beast. :-).  It is hard to describe, but it acts like a tree of gain control units, some aspects are sliding frequency are gain control, and rather than serial, they pool their efforts.  It took me a year thinking about it from time to time to understand a way of doing SR decoding without violating existing patents.   Everyone seems  to patent the easy way of doing things, but keep the good ways of doing things as a trade secret. On the dropped SR project, like on the DA --  'I thought differently'.

 

Thanks for the idea, but those various schematics are exactly the kind of thing that I would have loved to have availble before the FA side of things started.  Those schematics are DA though -- so don't really address these FA complexities.   It is good  to find every resource that you can before starting any project like this..

 

 

 

 

[John Dyson]

1 hour ago, KSTR said:

With this manual, the path is open to do a Spice sim with LTspice. LTspice can use .WAV files for time-domain input and output, so actual music files could be rendered, and then decoded and see how close one gets to the original.

Only problem is those JFET models and trimming their parameters so they match what Dolby had selected them for, though some educated guesses sure could be arrived at.

Even better scan here: https://www.richardhess.com/manuals/Dolby/Dolby CAT 22 Schematic scan 01.pdf

The JFET is one problem, but the selected diodes and resistors are also troublesome.   Cat22 schematic is good to start from, and also refer to R Dolby public papers on the subject.  They aren't very specific, but give some guidelines about what is going on.

 

Also, when starting projects like this or writing novel software, I always do patent searches.   Even though the Sony DolbyA patent(s) are interesting, and they do give a possible method for the EASY part of the decoder, that patent is not the best way to do it.   Their way is 'cool', and 'compact', but is a feedback design.   With a totally different feedback design, emulating another feedback design in a temporal (transient) sense become very tricky.   Different time and phase delay characters and alos with the constraint of feedback loops -- that would be total mess to try to make everything work very well.   With feed-forward, I have much better control, but also have to deal with a LOT more details, and some are very complex.  It isn't just about nonlinear functions, but some parameters need to be delayed before use, and that delay is most likly modified by gain, sqrt(gain), gain from another band (in the case of HF0/HF1), also dealing with the overlap of the HF0/HF1 compressors, recognizing that there is a delay through those, affect the result strangely, even in encoding mode.

 

I unfolded all of that stuff, measured actual DolbyA units for the curves.  Knowing the design, then fitting the curves (after some initial math tricks) and filled in some of the emulation parameters.  I couldn't get all of them, because like the new pre/de-emphasis on the FA decoder, parts of the charcteristics are not independent, so not easily measuered externally.

There was some 'tweaking by ear' on the DolbyA decoder to get everything right, and it ended working well.   This was TRUE TWEAKING and not what I am doing on FA.

My hearing  is definitely BAD on frequency balance, but I an hear a 'tell' from a signal from across the room, with my headphones volume turned all the way down.  (Exaggerating, of course.)   If I had ot do frequency response balance -- which I NEVER do, I make a mess of it.  I can NOT judge response balance, and given the results on the FA project, my hearing p problem is even worse than I thought.

 

(Oh, about matching DolbyA charcteristics -- I also have before and after decoding examples from 361, 301, cat22, etc.  Also have various commercial master tapes.  Also, I have recordings of strategically designed test tones and test signals, so it is much easier to do objective comparisons.)  Even with the test material, a lot of design did require subjective work.

 

 

[John Dyson]

The V2.2.6A decoder is ready, examples are ready, etc.

HOLD ON -- wait until tomorrow -- revelation!!!

 

You can read the stuff  below the main section here, but it is so optional that I probably shouldn't have wasted my limited effectiveness at typing and no written it at all.

These new results are truly plausible, but I do know the results are also not perfect. WE can make it closer to perfect.

 

I really like these results -- they generally sound good to me, and in these demos, in my opinion  I can hear only a few short instances where the FA has a more interesting sound chracteristc.  Even in my own case, the score will never be 100%, but 100% is impossible.  Refer to my 'Take Five' comments, even though I do vacillate between preferring the hissless version or the hissy version with more ambience.   Sometimes there is not a correct answer.

 

When I listen to the snippets, my software sequences through RAW/DECODED/RAW/DECODED...   So, I am supplying both, and also a directory with decoded only.  This is so you can simply sequence through decoded snippets only.  Normally, people only listen to RAW.  The DECODED only set of files gives an easier option to sequence through DECODED-only also.

 

As always, before critical comparisons, it will be more accurate to download first.   The Dropbox player tends to add its own distortions.

 

Snippets with RAW/DECODED, also the best place to get the latest decoder...

https://www.dropbox.com/sh/tepjnd01xawzscv/AAB08KiAo8IRtYiUXSHRwLMla?dl=0

 

DECODED only, if you want to avoid being biased by suspect FA quality (the same could be said about DECODED quality, but this gives you an easierchance to look it from a different direction:

https://www.dropbox.com/s/u9czgcpf38dkw3c/01 - Downtown-V2.2.6A-0-DEC-SNIP.flac?dl=0

 

FA decoder is in the RAW/DECODED directory, but here are direct links:

(It really is best just to go to the RAW/DECODED directory and look up the file -- there is NOTHING automatic about installing the decoder anyway.)

 

Windows V2.2.6A:

https://www.dropbox.com/s/yfz89egzgrmt1ic/da-release2.2.6A-win-2021-04-16.zip?dl=0

 

Linux V2.2.6A;

https://www.dropbox.com/s/n9qt84t3g77y29t/da-V2.2.6A-Linux.txz?dl=0

 

StartUsage document (probably should call it 'scribbles')

https://www.dropbox.com/s/1t42k430t0va1uy/StartUsage-V2.2.6A.pdf?dl=0

 

 

To make better progress

I need feedback on:

1) general response balance.  don't want to get mired in details on this matter, but does the response sound generally balanced.

2) about the highs -- are the highest highs  too strong or too weak.   This is where I wasted a lot of time this morning.  I thought the highs were dead, but instead it was my hearing.  When my hearing fixed itself, and I listened to the test results, I paniked this afternoon, thinking that it would 'nice' like yesterday, but turned out hideous.  I fixed a problem that didn't exist and made worse.

3) Lows are a little tricky.  Gotta do them correctly, even with the correct 2nd and 1st order EQ balance. I'd expect after 2  to 3 iterations, we can make this perfec.t

4) Do you hear dynamics problems?   Gotta be very specific about dynamics, especially.  I can hear most problems, but definitely

willing to find opinions on the matter.  I am not talking about wanting weaker/stronger cymbals.  It is more about surging/pumping/hiss modulation/etc.

 

I LOVE specifics...   I don't understand 'it sucks' and it only might hurt feelings with no benefit.  After the project is finished and working, then I'll be forgotten.  This is NOT about me, we should all try to do something good.

 

----------------

 

 

Here is a note about the 'bass' that was just changed, hopefully there would be an improvement that might even be made better.

--------------------------------------------------------

 

I permanently added the 4.5dB LF boost (probably should have been 5.25dB), and made the mods on the 1st order EQ so that the 2nd order 80Hz boost doesn't distort the response curve beyond where it should plausibly be. There might still be a problem, but since I cannot read the design engineers mind, I can only guess and use my experience and engineering capabilites to make plausible choices.

 

Even though replacing a small part of the 1st order EQ with an architecturally defensible 2nd order EQ, choices made by the original designer might have justified  some other design elements..  For example, maybe if the bass is really too thin, I might be able to justify a 3dB @ 80Hz as a -3dB @80 pre-emphsis before encoding to help mitigate the 'bass compression boost' done by the FA encoding process. I'll will think about this with a little more focus tonight and try to figure out an architectual ideal choice, but also try to 2nd guess what a designer might have actually chosen..  I need to make sure that everything  follows good engineering design practices and rules, or have good reason for exceptions.

Without specs, the FA decoder can never be 'canonically correct', but we can make it function correctly.   I was successful with incomplete information available when doing the the DolbyA project, but the FA information is A LOT MORE INCOMPLETE!!!  And, even worse -- my hearing must have degraded in the last few years.   With this realiztion, I can understand portions of the previous criticisms.   However, any representation that the old single layer design was better?!??!   I don't know whether to uncontrollably laugh or be personally insulted :-).

 

 

 

[John Dyson]

I  know that people really do have to be patient when dealing with me and this project, but the problems this morning diverted

some review about the bass.

The bass bothers me.   Even though the sound is good IMO, there is something wrong.

 

Just now, I went for the 'full monty', and used the largest justifyable bass EQ at 80Hz.   The test  results are making me think that I should have

been more patient about the release.   I'll do another release (I promise) tomorrow morning (+14Hrs..)   Methinks that

the bass is really too thin.

 

[fas42]

Something else to throw in the equation, John ... you're getting closer - by the time you finish, the DEC will sound identical to the RAW, 🤣.

 

But being serious, on the Mamma Mia track, on the "Just one more thing" crescendos, the sound is congested - it sounds harsh and closed, compared to the RAW ... hope this helps, 🙂.

 

Can't help on the bass, sorry - zero output below 200Hz on this setup ...

[John Dyson]

49 minutes ago, fas42 said:

Something else to throw in the equation, John ... you're getting closer - by the time you finish, the DEC will sound identical to the RAW, 🤣.

 

But being serious, on the Mamma Mia track, on the "Just one more thing" crescendos, the sound is congested - it sounds harsh and closed, compared to the RAW ... hope this helps, 🙂.

 

Can't help on the bass, sorry - zero output below 200Hz on this setup ...

Thanks for telling me -- I really am trying.

(The evidence has been piling up about the bass, just wanted to let you know that you really did contribute by the bass comment.)

 

I admit that the bass isn't good.  As you probably know, I am very bass adverse.  This is probably contrary to most peoples feelings about the matter, so I must try to compensate, and I really am trying. (Even though I talk about the rules -- at certain points in the design, there is a lot of flexibility, but must be constrained by good engineering practice.)  So, theoretically, I could add 3 +9dB EQ at 18.75Hz, and would follow the rules, but on the decoder would probably be very poor, insane engineering practice.

 

Yes -- they should definitely sound similar, and we are getting REALLY close to correct.   The decoded should sound more velvet (not prejudicially good), and more subtle.   The FA is a more busy sound.  If the FA encoding is used for one of the reasons that I think -- it might be for making the sound more 'active'.  Some people like that kind of thing.  It doesn't sound like that it the studio -- think more about the 'velvet' sound.  Even though in the studio, the dynamics are the most accurate that they will ever be, sometimes you might prefer a more 'dynamic' sounding recording.

 

There are a couple improvements that I now expect from FA to decoded.  I have modified the goals, because I now better know what to expect.   I knew about the modulation distortions and hiss, but a little over shooting the goal for vocals.  I am still not 100% sure what to expect about vocals because of the bass issue.

 

1) Less buzz & grain in the sound.

2) Less  compressed ambience, but as you might have noted, in my opinion, sometimes the noisy ambience can once in a very long while make a recording subjectively sound better.  Most of the time for me, the noisy ambience ruins a recording.  Take Five is an odd beast, and my preference changes from day to day.

3) Instead of the telephone vocal improvement, there is a less profound improvement in impairments, but still does good things like mitigating the worst of the artificial envelope inter-modulation distortions.

4) For when the decoded results are made correct, the bass has less of the distortion created by bending the waveform by gain control.   This might be two-edged for some people, where choice is a matter of personal preference.   The FA bass compression oddly makes it sound more 'dynamic' because it lasts longer and is kept louder longer  (because of the compression), and it has the modulation effects as a 'ring' around the envelope.  Normally, you'd think that expansion recovers 'dynamics', and that is mostly true.  Being 'dynamic' sounding and recovering the 'dynamics' on the recording can be two, very independent things.  In the case of FA bass, the FA sound IS more dynamic, but the dynamics aren't very accurate.  Which do you want?  'Accuracy' or 'Enjoyment'.   For some people, those are one-in-the-same.  For other people, maybe not...

 

I am working very conscientiously on the bass improvement. I cannot just add bass.  Anything that I do must be defensible.

I want the FA decoder to have integrity in the design, and also I am hoping that the decoder will be the best that it can be.

If I don't 'follow the rules', and not very careful, this thing can very quickly blow-up into chaos.

If there was a spec, then there would be some freedom to 'wing-it'.  On this thing, there is no freedom to do anything to the design that cannot be justified and/or defended.

 

Basically, given the limitations that we have, I am trying with the best intentions and the best integrity, to make the decoder as good as possible.  Also, I am trying to get a little fulfilment and small amount of enjoyment.  Sometimes one or the other is lacking for too long.  I hope that we have gone beyond the crossroads, headed for actual success, rather than being deluded about success.  Obviously, I have sometimes been deluded because of enthusiasm and strongly motivate  to do 'good works' for people.

 

 

 

 

 

[John Dyson]

I deeply apologize about this delay.   I know that this hasn't been a very stable, ordered development.  There was no basic design concept to start with, and after 5+yrs, I am still guessing.

 

I gave up on making the release available tonight.

The thing that @fas42publically wrote about the decoder output sounding more and more similar to the FA signal, and in some ways, that is true.

His comments about the bass were also true, and I am trying to 'DO THE RIGHT THING' instead of hack a solution.   I think that we'd all prefer the best

solution rather than something that 'sort of' works.

 

On FA decoding  result is, in colloquial terms -- it should be more 'velvet', because the crazy morass of noise and modulation distortions are removed.   If I originally  thought this would be all the improvement that we get, then I might not have started the project.  However, it was a challenge, and a LOT of people told me that the impairments didn't exist.  I even got pushback.  Noww e all know that the impairments DO exist.  That is important in some ways, but less important than enjoying your music. , Enjoying your music shouldn't be much about technical issues at all.

 

1)  we are so close to completion, might as well finish it.

2)  the sound is generally better than previous decoder versions, and IMO better than FA , especially on headphones where the details are so very obvious.

3) On headphones, very few recordings sound good to me, until being decoded.  I am fairly sure that there are other people who will feel the same.

4) Truly, the only recent problems with the decoder has been relatively simple EQ choices, that if I had good hearing, would not have been troublesome at all.

 

It is pretty obvious that most of the previous troubles on the project have been related to errors that I make based on information developed from bad hearing.

Please continue to be patient for a few more releases.
 

Even though I am trying as diligently as possible to make the decoder fully functional, I suspect my hearing is getting worse.   The insanity this morning shows that I must get this thing done quickly -- days/weeks instead of months.   Off topic, but yet another reason to go to the doctor...  YUCK!!!

 

I won't make a promise, but I'll be working essentially all night to make the program available that I had intended to release a few hours ago.  You know my usual time:  9:00AM, USA Eastern time.  A little more than 12Hrs from now.

 

As of right now, since the botched release attempt,  I got help with input from a few kind people, and I did NOT respond with expletives, but tried to make sure that I understood what was reported.

 

 

[fas42]

51 minutes ago, John Dyson said:

 

 

I gave up on making the release available tonight.

The thing that @fas42publically wrote about the decoder output sounding more and more similar to the FA signal, and in some ways, that is true.

His comments about the bass were also true, and I am trying to 'DO THE RIGHT THING' instead of hack a solution.   I think that we'd all prefer the best

solution rather than something that 'sort of' works.

 

 

 

Don't get confused about what I'm saying about bass, John - my laptop speakers are incapable doing any frequencies below 200Hz, irrespective of what they are fed. Which means that I'll be unable to comment on any aspects of the bass adjustments that you do - okay?

[John Dyson]

1 hour ago, fas42 said:

 

Don't get confused about what I'm saying about bass, John - my laptop speakers are incapable doing any frequencies below 200Hz, irrespective of what they are fed. Which means that I'll be unable to comment on any aspects of the bass adjustments that you do - okay?

Honestly, I have been having trouble with the bass for months...  I got the basics working now-- previous was REALLY STUPID thinking with blinders on.  Next, how do we the correct amount based on the lower frequencies of the decoder working correctly now..

 

ADD-ON:  Just now, after a lot of testing, I found a 'tell' for the LF EQ, and optimized it.  It turns out to be a very rational result.  The effect is to open 'Anne Murray's vocals while still having a strong, natural bass, but not quite at the FA level, but very reasonably close.  Her voice is very well centered an has a 'space' between the bass sources from where it emanates clearly.   No serious mud at all.

We have had two obsticles, and I believe that I have been both of them.  Anyway, there was some bass below 200Hz, but definitely insufficent.  It took me untiil tonight that I now understand that the bass was wrong -- not just wrong settings, but wrong kind of EQ.  (You know, on input/output, EQ is a requirment in this design.)  This kind of EQ is part of the design, not a listener preference.

 

Anyway -- this is a really really long story, but the bottom line right now, I have the correct pieces, and have them work *plausibly* correctly now.   There is now a reasonable approximation of the strong FA bass -- so it is close.  The problem now is, what should decoded bass sound like.  We can make the levels correct, but if we extend the bass too far down with 1st order EQ, I promise you -- you will not like sound.   There is really a needed shape to the EQ, flat doesn't work.

 

There is a tradeoff of this evil kind of bass which is more flat, but sounds like air pressure against you ears, not like a drum or bass guitar.  Actually it is kind of weird.

So, I use 'the rules' to respectfully shape the response, using a mix of 1st order and 2nd order EQ.  It is like a Tetris game now.  I have pretty much gotten the

2nd order EQ correct -- but 2nd order EQ is easy to tune by listening.   1st order EQ, if you are trying to match an encoding process, can drive someone totally crazy.  I think that part of the problem for me, 1st order EQ is so soft in the response curve, it doesn't general 'tells' that make any sense.  Also, when you use say, 6dB at 20Hz EQ, that is really wide, and will have signifcant effects perhaps up to about 80Hz.   If you use 1.5dB * 4, it is still 6dB, but it only affects frequencies up to, maybe 35Hz.   If one knows that they need -6dB at a given frequency, it usually isn't correct to use -6dB, but instead it usually appears that -3dB * 2 works better.   This can really make you crazy.

 

We are talking about one frequency when discussing this.  For the LF EQ that  I am trying to make it work correctly, there is 2nd order at 75Hz (finished), which ended up being 3 earch of 75Hz low pass shelf, -3dB, Q=1.17.)   I am not 100% sure about the Q value, but I know that I have it close.   For the 1st order EQ, we don't just have 1EQ,we have a lot of them  -- we have 75Hz, 50Hz, 37.5Hz, 18.75Hz, and probably 9.375Hz should also be done correctly.   Those are lots of EQ settings, if we need 3dB, which IS correct for most of them, do we really need 1.5dB + 1.5dB, or do we need 3dB?

 

It just so happens that it sounds like +3dB at each of 75, 50, 37.5 is too much bass.   Which one do we disable?   Do we disable the 75Hz one, or we disable the 37.5Hz one.  Most likely 50Hz should not be disabled.   What sounds correct?   Can you imagine me making a descision like that?  Can you imagine how inaccurate my answer will be?  Maybe even all three having +3dB is correct -- but my judgement of 'response balance'  is about as accurate as random chance.

 

Right now, I am VERY CAREFULY trying to compare sound, attempting to get most of the dynamics of the FA sound, most of the general levels -- but YOU KNOW that the bass WILL sound different, but what KIND of different is correct?

 

I WILL do the best I can for tomorrow morning, and after the release, hopefully get some useful criticism trying to tell me what the listener feels like the defects are.  If there is enough constructive feedback, and I read their reports very carefully, we might just finish this project very soon.

 

Thanks for writing about this with me -- this darned thing is really close.  In a way, I think that our old friend is helping me -- I do think about him.

 

[PeterSt]

Do you really expect anyone to read this all ?

[semente]

Hi @John Dyson,

 

Sorry if this has been asked before but would it be possible for you to list all the steps that you take when you are performing "de-emphasis/decoding" of feral A albums editions/masterings?

 

My second question may become irrelevant after you reply to the first but I will ask anyway: are you performing only "decoding" or are you on top of that trying to "remaster" the tracks in order to perceptually improve their sound?

 

It's been some time since I last followed your work for mutiple reasons, mostly because I don't listen to most of the music that you've chosen to "decode", but there are albums which I still sometimes listen which I believe could benefit from the "treatment".

Also because all my computers are Mac and I cannot use the programme that you've kindly made available to the community.

 

One more question that's just come up to my mind, possibly a dumb one.

Wikipedia describes Dolby A steps as follows:

The input signal is split into frequency bands by four filters with 12 dB per octave slopes, with cutoff frequencies (3 dB down points) as follows: low-pass at 80 Hz; band-pass from 80 Hz to 3 kHz; a high-pass from 3 kHz; and another high-pass at 9 kHz. The compander circuit has a threshold of −40 dB, with a ratio of 2:1 for a compression/expansion of 10 dB.

Are feral A albums the result of recording masters without Dolby A pre-coding which went through Dolby A de-decoding process during mastering, thus explaining the exaggerated treble and limited bottom end?

 

master tape with no Dolby A pre-coding → mastered with Dolby A de-coding → CD

 

Thanks,

Ricardo

[John Dyson]

V2.2.6C is all uploaded, and the results for simple decoding are starting to be as good as they are likely to be (within reason.)  I (we all) didn't know what to expect, and I think that I oversold some of it, but decoding is a technical improvement for most material.  Noting the comments below, some people might not prefer one or the other part of FA decoding, but the goal of the decoder is oriented towards the 'technical' aspects, not directly focused on listening enjoyment.  It is very likely the decoder will improve listening enjoyment for those feeling that technical quality is most important, even though the 'RAW' material sound will also be more enjoyable to many people.  Sometimes, people like me will find that SOMETIMES the RAW sound might be more enjoyable, however technically more flawed.   At least, when I put my audiophile hat on, the pure technical quality is a lower-priority first-level-attribute than the music itself.  My 'engineering' hat has been on for ever since I joined this forum -- being very very frustrated by my hearing and lack of ANY specification.  Both of these challenges make success almost impossible.  However, I don't fail, as a mathematical truth. :-).

 

 

https://www.dropbox.com/sh/i6jccfopoi93s05/AAAZYvdR5co3-d1OM7v0BxWja?dl=0

 

After all of this work, and the necessary interactions,  I have learned some technical things, some stuff about human nature..., but also have something to share: consider focusing on listening.  I have surprisingly noticed a lot of  the 'yet another' styles of projects...  Myself, I get terribly bored with 'yet another' projects, so took on the FA decoder (and originally the previously impossible DA decoder) as challenges.   I have turned down some 'yet another' projects -- 'yet another' projects  don't get my adrenaline going.  Back when I had to work, I got paid to do 'yet another' projects from time to time.  Now I have the freedom to truly do new things. 'yet another' type projects are "work",  really new stuff is more enjoyable for me.  What if more of the technically oriented audiophile energy from 'yet another' projects was spent instead on doing new stuff?  I wonder where we (the technologically dependent world) could be today...

 

Back to the project, I have found another  kind of distortion in the FA signal, 'phase scrambling'.   The decoder has some 1st level phase descramblers in the code, and is probably a major reason why 'Take a Chance on Me' was clarified in some of the pre-V2.2.6 versions.   I have disabled the 'phase descrambler' portion of the decoder, and using only the minimum 'special sauce' to mitigate dynamics processing damage.   I am moving forward to look at phase descrambling, because it is very likely to make yet another performance/clarification improvement.  This 'phase descrambling' was  a portion of the improvement in some of ealrier demos (yes, I know -- the earlier demos weren't necessarily all that great, but the phase descrambling really did help.)  Some initial attempts at phase descrambling in the decoder code have been mostly successful.  Since it is becoming obvious to me that phase scrambing has been done to the signal, I can focus on the problem better.

 

The decoding result shows that the DA chain complex alone isn't all of the damage being done by FA encoding, but is most of it.  The DA chain is pretty close to complete.  I will probably spend a few weeks on the phase descrambling and code/algorithm improvement for the project in general.

 

The damage apparent in the 'far field;' as created by FA, is  more weak than I had originally thought, but also decoding the material can have a profound effect towards attaing truer 'high fidelity'.   Decoding is less important on average for the general audiophile -- just those who are more focused on the detailed technical aspects and those who have very good coupling between their hearing and their systems.   Perhaps headphone audiophiles should strongly consider decoding material simply because it DOES remove a lot of HF dynamics distortion and hiss that is so very audible when using headphones.

 

Decoding material can very significantly improves quality, if low level distortion, transients (e.g. cymbals), hiss, and clarity are important to you.  Do NOT expect the profound improvement that I thought that there would be and proclaimed the possibility, but for headphone listening, doing an FA decode can make an amazing audible  improvement.  It will not always make an amazing improvement in clarity and removing 'fuzz', but even in the most benign case of FA encoding, decoding it will likely give  SOME improvement in the HF part of the signal.

 

The matters  become even trickier/more complex at the LF (I don't mean DolbyA LF, but I mean 200Hz and lower.)   The decoding operation DOES profoundly change the sound of the fidelity aspects of LF.  Decoding recordings with significant LF will remove a lot of the technical aspect of LF distortion.  Whether or not removing that technical aspect of distortion is desirable or not would be based on personal preference.   There is be a place for correcting the technically distorted bass. It is also likely that a lot of people might rightfully prefer the technically distorted bass.  'Distortion' is not always bad, ask any recording engineer.

 

We are getting pretty close, and once this thing settles down, I'll start the last phases of the project.   Those phases will mostly be less audiophile oriented, but still important to do.

The FA decoder part of the project has been important, but some of the design concepts in the DA decoder are likely more useful techniques for other projects.