'FeralA' decoder -- free-to-use

[Jud]

6 minutes ago, John Dyson said:

I meant, the consumer expansion devices intended for the dissatisfied HiFi consumer like 3BX or somesuch. 

Possibly quite wrong about this because I'm basing it only on my personal anecdotal experience, but perhaps (lack of) marketing, because I was only very dimly aware of anything *bx and thought of it as pro stuff. Dolby pretty well dominated consumer compander mindshare for me.

[John Dyson]

47 minutes ago, Jud said:

Possibly quite wrong about this because I'm basing it only on my personal anecdotal experience, but perhaps (lack of) marketing, because I was only very dimly aware of anything *bx and thought of it as pro stuff. Dolby pretty well dominated consumer compander mindshare for me.

I didn't mean to be 'know-it-all' -- it has been many years since the 1980's.   Things were very different, where 'tone controls' were stiill the norm, but on their way out.

 

I have been very surprised about the extreme precision used nowadays and my assumptions about precsion create lots of the mistakes during the decoder development.   The first lab versions 7 or 8 yrs ago of the DolbyA decoder were attrocious.   The 'improved' DolbyA deocder used expertly optimized RMS detectors for level measurement.   Using a perfect detector with minimum ripple was exactly NOT what a DolbyA decoder needs.   R Dolby already designed a fantastic scheme, and his detectors were 'perfect' in a different way. I have made many mistakes, and learned A LOT.

 

The 3BX/4BX/5BX device was a multi-band expander whose purpose was to try to improve the dynamics of consumer recordings.   There were also devices more similar to the Dolby approach available from manufacturers like Pioneer.   These devices would push hiss downwards, hopefully to inaudibility.  Also, the peaks would be slightly improved.  The Dolby approach tends to be a carefully crafted attack/release scheme over a narrow gain range, but DBX stuff is more like an optimum detector that processes all of the dynamics over a wide range.

 

I have learned, through lots of trial and error, that general purpose expanders are like a swiss army knife with all of the knives fully open.   That is -- the consumer expanders could sometimes improve the sound of a recording, but more often than not will damage the sound.   The 5BX could theoretically get rid of huge amounts of hiss, but would very often add artifacts that make the user wish for the good old days of hissy recordings with limited dynamics.

 

The only way to clean up consumer recordings 'correctly' is to undo the IP (intellectual property) protection process, and that is what the FA decoder is about.

No-one knows everything.   It seems like I know less and less though the years.

 

 

 

 

[John Dyson]

Sorry for the delay (about 1 day), but after mentioning some attributes about the descrambler, esp about how much expansion it can do, I have decided to use a higher setting for the descrambler.   This higher setting requires equivalent compensatory settings elsewhere, and without a spec or more test materials, this new try for the correct settings is as good as any.

 

The new setting sounds reasonably good, is dynamic and is very clear/clean.   One of the previous versions was a little too 'dirty' or 'compressed'.   This 'dirty' sound came from the descrambler being set on the low side of correct.   This new try is on the high or 'strong' side of correct.

 

It will take a little more time to complete and verify the changes being fully made, and then a few more hours to do the decodes.

Frankly, I like this upcoming version better, but is also uncomfortable cause of the strong setting.   Even though the descrambler is frustrating to work with, it is also amazing.

 

I am expecting that the upcoming try will be 'anti-FA' sound character.

 

 

 

[John Dyson]

Still working on finalizing the next version, and it will be similar to X8.1EP21 for <response balance>, but that version has some other defects.   There have been over 40 test versions since EP21, and the best of all test versions are being merged.   EP21 is the best for <response balance>, but the descrambler fine-tuning and gain settings were incorrect.   Also, there was an important part of 9kHz rolloff that was wrong, and the anti-distortion was mistakenly enabled.   I have decided that most of the anti-distortion was a continuation of a good idea too far into being incorrect.

 

It makes a lot of technical sense why the response balance might be close to correct on 'EP21', the EQ really does fit a pattern consistent with what is needed to mesh with the descrambler.   The dynamics and some of the imaging (anti-distortion) isn't correct in 'EP21', and also there were complaints about the low level 'flutter'.   The 'flutter' issue is noted and is actually 100% understood.  The 'flutter' comes from an unhappy situation about the EQ in the FA encoding process where *at low levels* the LF controls the HF gain more than the HF does!!!  You won't normally notice that fact in playback, but when looking into the low level code,  understanding how the EQ must function, the matter where the LF drives the HF gain becomes evident.   WIthout access to the LF, the low level fade-in HF will wobble all over the place...   Perhaps it would be most accurate to describe the LF effect on the HF EQ is that the LF becomes a 'bias' for the HF gain level.   Then, the HF makes the minor gain changes to itself.   For extreme situations, FA is basically a cluster '****' to decode!!!!   The descrambler is also a tricky thing, but EP21 is a manifestation of the base, deep-level descrambler apparently being really correctly functioning.  There are other aspects of EP21 that were totally incorrect.

 

No matter my excuses about the 'flutter', it is always considered to be important, but sometimes missed on these test releases.  The next 'X8.1EP' version will be a test version/release, and WILL have consideration for the 'flutter' this time, but even after due considerastion, please don't be disappointed about the flutter.   The pure LF driven perturbations can never be 100% corrected, but the modulation of the HF should be mostly mitigated.   Criticism about the 'flutter' is still welcomed, and will tend to force more innovation in that area.   Previous criticism, however condemning, has forced continued improvements.

 

* Also, I did find a bug in the DA decoder, where the level is not as stable as it should be.  The feedback is still a little too sloppy with too much time delay.  The DA decoder in the upcoming FA decoder version is improved, but will be focused on in the next day or two.   I don't know if the DA correction is even noticeable in FA decoding, but no matter what it needs to be improved.

 

PS:  The first attempted DA fix does improve the 'flutter' markedly.   It makes the low level feedback loop more *accurately* responsive, therefore the small undulations in the LF signal are better tracked.

 

 

 

 

[John Dyson]

Potentially wonderful news about the next version.   There were bugs in the descrambler where the list of frequencies had jumps in it.  The unevenness causes distortion like effects.  Frustratingly, my brain will 'fill in' the differences so everything looked correct.   The list is fairly long and repetitious (every freq step from 1.5kHz on up to 31.5kHz in 1.5kHz steps.)

 

With the new bugfixes, I am working on a release, and also going to provide the decoding results to an external reviewer.   The initial demo showed a very positive response.

Today's decoder is *not* yestereday's decoder.  I am working on bugs that were previously 'little nits'.   The corrections are becoming very very fine, previously ignored.

 

This work has been time consuming because the corrections are not one or two major ones, but instead perhaps 20-30 very minor investigations/verifications and/or bugfixes.   The 'list' is exhausted, with the previous decoder version having 2 major bugs -- bad LF setting and descrambler now working perfect.   The promised stereo image now fully fills the soundfield.

I know that I have mentioned this in the past -- the descrambler is a very difficult to figure out.   The main thing is that there is some auxiliary EQ needed along with the standard descrambler structure.   The additional EQ sets the dynamics for HF (makes the higher HF have softer dynamics to match the lower freqs.)   Once the HF matched the MF better, the stereo image got better.

 

The final V8.1EP might be close to ready.   This version is also going to a significant non-AS related reviewer.  I have been focused on the cleanup more strongly than ever.   The test version might be existing now, or if another  minor bug comes up -- then it will need to be corrected.

 

 

[Audiot]

Hallelujaudio ✓ 

[John Dyson]

Just made some preiminary copies of some recordings, the eventual versions will be for external, higher profile, review.  (Definitely circumspect, huh?)

The descrambler is a magical beast -- I am still confounded by how much processing that such a simple 'circuit' can do.   The magical beast is being 'tamed', finally doing some really impressive 'tricks'.

 

I'll be making snippets available publically tomorrow (CD1 of Olivia Newton John's 48 singles.)   For the final public review version, there are still a few very minor issues that need to be 'looked at'.   I don't know if the 'issues' are 'problems' or simply characterstic of the recordings tested.  There is further processing needed to make snippets,will make them publically available tomorrow,  and frankly I need to sleep now.

 

*  Some of the tracks on the ONJ recordings don't sound like what I think that they should.   It is likely a descrambler setting issue.   Still working on it, and the reason why the demos are preliminary.   Easily within a day or two of being releasable though.

 

Do not expect the sound from older versions of the decoder.   A lot *I mean a LOT* of things have 'popped' in the last week or so.   My 'quiet' online is because of the ongoing improvements  in reverse engineering accuracy.

 

Being very straightforward -- the decoder is being 'zeroed-in', with progressive interacting improvement.   A few hours ago, a sweet spot in the decoder settings has been found.   These settings are true 'settings' as chosen by the original designer, and they must be exactly the same to get the optimum 'pop'.  There are approx 10 active BINARY settings right now, but there is a  pool of probably another 20 settings that appear to be correct.   There is still some 'wobble', and it is about how the HF descrambler dynamics should be 'pushed' by the LF.   The interactions with the descrambler are much more nightmarish than I had even imagined 1wk ago.   The word that comes to mind starts with a 'c' and ends with 'ck'.

 

When/if you decide to hear the demos tomorrow, you will understand what I mean by 'pop'.   A few correspondents here on AS have pointers to the recordings, and might chime in after listening.   The decoder isn't quite where I want it to be, but there is strong external motivation to make it 'reviewable' in a more formal sense in a few more days.

 

 

 

[John Dyson]

About 24 to 48Hrs early, happily the final experimental X8.1EP release has been created, saved off, etc.   The decodes are starting.   No current private demos are now up to date, and I'll be updating everything before 24Hrs.   There will be a decoder available this weekend.  This one sounds very, very different -- much more natural, and much better controlled highs.

 

Previously, my hearing was the only guide.   However, the 'keyhole' of the basic original design was found a week or so ago.   That was when I became quiet.  The 'keyhole' still had lots of variables, the most challenging was which module should be part of the descrambler complex, or instead before or after.  Since the descrambler is parametrically nonlinear, depending on where the modules reside, the descrambler can act very differently.   Honestly, there was one bug that was wrapped up yesterday, and it was the precisely correct EQ style for the LF pre/post emphasis.   The frequencies and gains have been correct for a very long time, but Q values and EQ structure make a difference, but not as profound as the correct choice of frequency.

 

The 'keyhole' gave significant guidance about the needed HF EQ, but still didn't give the precise values.   There was still the inside or after EQ choice.   The 'before' EQ choice appears to be directly related to the inter-layer needs.

 

I know that a lot of this sounds like gobblty-gook, but it is technically accurate.   The parametric nonlinearity aspects of the descrambler still need to be better understood by me.  The amount of expanding available can be very overwhelming, but the correct choice is when the HF melds into the rest of the sound.

 

Unless health emergency or severe family distraction, the decoder and demos will be available in 24Hrs.   I am listening to the results as they come from the decoder, very good sound...  Unlike even the first private examples, there is NO garble in anything.

 

John

 

[John Dyson]

Since there was discussion earlier about 'Katy Lied' (at least private discussion), and mention of the strange mastering, it seemed to be appropriate to attach a decoding snippet using X8.1EP-REL.   The matter of the missing tamborine seems to have been resolved, however suppressed it seems.   The 'tamborine' doesn't move around or change in amplitude, that being a good indication of the possibilty of accurate FA decoding.   The 'Katy Lied' example has NOT been a target for the decoder internal configuration, and just happened to work farily well.

Other materials were used for X8.1EP test reference, some are pretty darned difficult to decode.   Still a bit of a challenge:  'Top of the World' by the Carpenters.  Instead of being intolerable sibilance, the anti-sibiliance cancellation now provides a more tolerable excess sibilance.    Other materials do tend to provide 'sweet' sibilance instead of 'crushed', but there is still a lot of material with 'crushed' sibilance, might be an attribute of the recording.

 

Katy Lied snippet using todays (<24Hrs) upcoming X8.1EP release (yes, release), and will eventually result in a V8.1EP release perhaps in a month or so.

 

Katy lied:

Katy Lied-Rose Darling.flac

 

[John Dyson]

8 hours ago, John Dyson said:

Since there was discussion earlier about 'Katy Lied' (at least private discussion), and mention of the strange mastering, it seemed to be appropriate to attach a decoding snippet using X8.1EP-REL.   The matter of the missing tamborine seems to have been resolved, however suppressed it seems.   The 'tamborine' doesn't move around or change in amplitude, that being a good indication of the possibilty of accurate FA decoding.   The 'Katy Lied' example has NOT been a target for the decoder internal configuration, and just happened to work farily well.

Other materials were used for X8.1EP test reference, some are pretty darned difficult to decode.   Still a bit of a challenge:  'Top of the World' by the Carpenters.  Instead of being intolerable sibilance, the anti-sibiliance cancellation now provides a more tolerable excess sibilance.    Other materials do tend to provide 'sweet' sibilance instead of 'crushed', but there is still a lot of material with 'crushed' sibilance, might be an attribute of the recording.

 

Katy Lied snippet using todays (<24Hrs) upcoming X8.1EP release (yes, release), and will eventually result in a V8.1EP release perhaps in a month or so.

 

Katy lied:

Katy Lied-Rose Darling.flac 2.01 MB · 4 downloads

 

IMPORTANT:   the quality of the actual release (sorry, I guess I found a few more improvements) is MUCH MUCH greater than the Katy Lied example.  Sadly, Katy Lied itself isn't much better  -- really regret it.   It would have been WONDERFUL to clean-up the recording (Katy Lied).   The 'decoder' only does 'decoder' things thgouh.   It might be possible to use the dispersive expander/compressor on a one-off or a general purpose processor, but the FA decoder is beyond it's abilities on this one.

 

For the release, true release, several hour delay -- but really tonight/this morning (USA EST time.)   REALLY sorry about the delay, but found another missed 2.213 ratio, mistakenly used 2.5 instead.   Believe it or not, this ratio is related to the stereo width conversions.   This minor change IS minor, but at this late date takes LOTS Of time to test/verify.   Also, had a personal errand that took up a lot of time -- but promises are promises, but I couldn't follow through.

 

The 2.213 magic number is all over the place -- like a secret encryption key.   For now, there are about 5 places where the magic number is explicitly defined, and of course, multiple definitions need to go away -- should only be one.   All kinds of variations of the magic number are used, e.g.  sqrt(magic), sqrt(sqrt(magic)), etc.

 

I really fully expect the release tonight.   There have been several full decoding runs today, and there were NO unsatisfactory decoding results today -- everything beautiful, clear, etc.   The 2.213 fix today is only obvious in certain, very few, recordings.   Never audible in any classical that I have, only audible on intense pop.

 

The decoder must be as perfect as I can make it.   Truly, there are few other places where it can be corrected.   Some day, I'll document the descrambler -- amazing expansion capability.

 

SORRY AGAIN -- fully expect something in about 10Hrs.   Full effort is being made.  Only slight changes, but need to be tested.

 

John

 

[John Dyson]

Deepest deepest apologies.   I just cannot work any more tonight.   It is important to get this release done, but there are too many details, too much testing, too much review to make a reliable release tonight.   I didn't realize that the minor correction for the stereo image to use the 2.213 base value goes into the fundamental calculations of certain critical things.   I thought that it would be a slam-dunk.  Instead, there is a chain of dependencies that need to be reviewed.   It is unwise to promise a time anymore, but very probably tomorrow.   Too much hurrying causes too much 'wobble' like in various premature 'demos'.   I REALLY want to share, but premature sharing produces too much embarrassment.    It is best just to finish instead of 'tease' with defective test versions.   (Changing the number rippled through the decoder and caused a serious EQ problem...   Whoa...    Really uprising, but also opened up a few other things to test.)

 

Believe me, the decoder is NOT something that a normal person can keep all together in their minds.  The details, structures, algorithms in every place in the decoder are probably impossible to remember except for the most brilliant high IQ person (not me.)  Exuberance often seduces the programmer (me) into making bad time estimates, therefore misleading the interested reviewers/users/well-wishers, etc.    The descrambler and DolbyA emulations are of similar challenge, but with different technical disciplines.    I sure wish I had someone to talk to about this stuff in person, but being an advanced audiophile doesn't necessarily prepare a person for the level of technology in the decoder.    A few people have sometimes tried to communicate technically, but the context is mind-bending also.  

 

For the several people who might be looking forward this beautiful thing, I really need to beg forgiveness.  The reviewer heard a defective, but actually very good sounding version.   That version was basically 'okay', but unable to handle the difficult kind of material that is being tested right now.   At least, the reviewed output makes a good baseline for comparison  using my sometimes unreliably deaf ears -- that fact has been helpful in the last few days.   The release is coming just as soon as the exhaustion is remedied.

 

I know that I keep on promising -- I am definitely wanting a debugged decoder than anyone else in the world.   Today, tomorrow, maybe day after -- it will be ready...  It probably IS here.  

 

We can see the spaceship a few feet away,  moving towards us, but still not quite docked yet.   I think that we are starting the process of 'docking', not much more travel.  Mostly must make sure that everything is lined up.

 

Time to sleep for at least a few hours.

John

 

 

 

[John Dyson]

The goal for release is Friday night, USA EST, but might be Saturday.   Absolutely every little nit is being reviewed over and over again.   I know this thing can be perfect.   It is a mindnumbingly complex monster, yet well designed.  Part of previous problems were 'marathon development' and 'testing' sessions that resulted in insane A/B comparisons (hearing problems).   Being much more patient than ever before.   We all know the goal, but now,  it is time to 'do it right'.   If there is a delay beyond Friday, it is only for the best reasons.

 

John

 

[John Dyson]

Status:

The sound is still a little bit 'off'.   Working on it, all of the 'pieces' are mostly correct, and as always, just a matter of judging the correct combination.

Currently, the result is a bit metallic sounding in the '3kHz to 6kHz' range.   It is 'bright ' sounding, but not 'burning'.   There is no excess >9kHz anymore.   The decoded sound has similarities with FA, much more clean, but still a bit too much middle HF, as if there is still a peak in the 6kHz range.   Sibilance is very clean, but slightly excess.   Not nearly as excessive as earlier versions, but there is still a small error.   Even the most egregious recordings are now easily decoded, including ABBA Gold and 'Dreamworld'.

 

Just adding 'EQ' is worrisome because it is the same thing as adding 'hardware' back in the late 1970s'.   As such, it is unlikely to be the correct solution.   Nowadays, adding 'hardware emulation' is easy and very cheap, but back in the 1970's, EQ was expensive.   More than likely, the needed change is slight and more likely 'removal' of EQ steps instead of adding.

 

The current failed version was a release candidate -- one or two more test candidates are needed, each try takes 4-6Hrs.       More than likely, there will be no release tonight (<12Hrs), but very possibly tomorrow night (24->36Hrs.)   (I use relative time because not everyone is in USA EST time.)

 

I regret the seemingly slow progress, but the code is being simplified with improvment in smoothness.   Simpler is better, more likely more correct, because HW was expensive in the late 1970s'.

 

John

 

 

[Audiot]

My friend said that a common assumption is that; "too much energy in the 8kHz range" is due to an EQ bump in that very range - when it is sometimes more likely because that too much has been added in the ~4kHz part of the spectrum... and.. overtones? 

With this example in mind (which is what he said, real life example)(and what I wrote you in PM regarding encoders and decoders, channels et. al) the "/ 2" factor could explain it. So if you hear excessiveness in 10kHz, try reducing the 10kHz / 2 area :)

[John Dyson]

8 minutes ago, Audiot said:

My friend said that a common assumption is that; "too much energy in the 8kHz range" is due to an EQ bump in that very range - when it is sometimes more likely because that too much has been added in the ~4kHz part of the spectrum... and.. overtones? 

With this example in mind (which is what he said, real life example)(and what I wrote you in PM regarding encoders and decoders, channels et. al) the "/ 2" factor could explain it. So if you hear excessiveness in 10kHz, try reducing the 10kHz / 2 area :)

The EQ system, actually the descrambler, works in steps of certain freqs, including  some freqs below 3kHz, 3kHz, 6kHz, 9kHz among others outside that range.   (There are two kinds of EQ, each at different intervals.)   The descrambler doesn't act like normal EQ, but suffice to say that the change to remedy the excess 6kHz region doesn't really change the general frequency response balance, but decreases the amount that the descrambler expands.   In many cases, the most recent change to remedy the excess 6kHz doesn't always produce an obvious difference in the sound.   The recent bug on test code is most noticeable for certain kinds of sibilance.  Without the signal variance, the descrambler bug is only slightly noticeable, but definitely not in response balance per se.  The two kinds of EQ freqs are below:

 

Main EQ sequence:  1.5kHz, 2kHz, 2.5kHz, 3kHz, 6kHz, 9kHz, 12kHz, 18kHz, 24kHz, 30kHz.

Secondary EQ sequence: 1.5kHz, 2kHz, 2.5kHz, 3kHz, 4.5kHz, 6kHz, 7.5kHz, 9kHz, 10.5kHz, 12kHz, 13.5kHz, 15kHz, 16.5kHz, 18kHz, 19.5kHz.... 30kHz.

 

Each of the frequencies in the secondary sequence have a special sequence of EQ that does the 'descrambling' action, and is intricately designed in a consistent way.

There are opportunties to add 'energy' at each of the main sequence frequencies, and to dissipate additional energy for expansion at the secondary frequencies.

 

When thinking about the descrambler, it is NOT the same as flat EQ, but instead it is more like 'energy conversion' that provides for signal expansion.

 

Dissipating the energy at any of the secondary frequencies (with priority given to the main sequence frequencies) effects normal 'EQ'.   Adding energy is done on a generally selective basis -- not all frequencies of the main sequence need additional energy.

 

Think of the change like this:  the descrambler expansion ratio was decreased in the 6kHz region, not that the response balance was changed.   The descrambler doesn't make a profound difference except for 'expansion' action.   It does not work on 'levels' but works more on 'changing levels'.   The responsiveness for the 'changing levels' in the 6kHz region has been decreased, therefore tamed the sound.

 

Without the 'descrambler', and implementing the EQ in a normal way produces a 'thin', irritating sound.,   The descrambler makes the sound more 'real'.

 

The initial result from the recently described change is very good, and if I was absolutely sure that the 6kHz correction fixed the last problem, I'd try to push for a release.  Instead, a full test run will be done starting now, and I'll review the results after staying away from the decoder/music for at least 3 Hrs.   After 3Hrs of 'decoding', I'll start reviewing the results.   I cannot judge the results during intensive testing & development, but instead 'giving it a rest' for several hours, doing evaluation after some rest.   There is NO NEED to try to verify the results during intensive development because the results are very close to correct.   In a static sense, the results ARE correct and have been close to correct for the last week.  The errors have been in the descrambler behavior.

 

John

 

 

 

[John Dyson]

Some context might be missing, but this is a reply sent to another interested party.   The discussion should be fairly obvious.   I am just too busy to put together another note that says the same thing :-).   As always, progress is being made, but there has been some very concrete feedback, and it is being carefully considered.

 

A version could be spun-off right now, and it might sound pretty good.   Right now, there are really no significant audible flaws except for accuracy.   'Sounds good' is probably now been achieved, but 'sounds good' is not enough.   I am desperately listening for 'tells' to find any 'hints' about accuracy and 'zeroing in'.   Some 'tells' were just found in the last day or so, with very positive results coming from the information....   Here is the note quoted from another discussion:

-----

Status report...
There have been some candidates, actually testing a very good one right now.
It might take one or two more candidates before the decoder is ready for release. Lots of comparisons to do.

However, when doing comparisons with the version that you tended to like, the sound is a little less concentrated in the lower HF (1.5kHz) region on the newer stuff. The newer result might be described as being similar to the sound of a graphic equalizer with a slight HF/LF boost. (The old 'smile' pattern, about a dB or so -- much less than typical graphic EQ settings though.) The earlier version, only in direct A/B comparisons, might seem to be slightly more nasal (not in a qualitative sense -- just a quantitative sense.) The sound on the new version has a bit more distinct highs and the bass has a bit less detail but is more smooth.
 

IMPORTANT: all of the EQ is based on the design, with no real tweaks. The comparisons are based ONLY on A/B and not 'sounds good'. The old criteria of 'sounds good' ends up being a mess, especially based on my evaluation of 'sounding good'.

I am trying to reconcile the 'new' version, the FA (original CD) version and the version that you *rightfully* prefer. I consider the version that you prefer as being equal in liklihood as the others'. The biggest challenge is my incredibly unreliable hearing, so it has slowed down progress a little. I cannot even do a reliable direct/immediate A/B comparison for response balance. This characteristic really slows things down!!!

There is currently a release candidate version preparing the demos right now. There might be one or two more candidates (about 24Hrs) before I"ll be comfortable. In the last few days, there have been a few minor 'breakthroughs' in finding sources for erroneous settings. The results are 'zeroing in', but until the current set of demos is carefully reviewed, I won't know if the decoder is ready or not.

Why is this taking so long? Trying to change the results from 'happenstance' to 'reasoned'. This has been a forever process over the last several years, and we are within days or hours of a pure 'reasoned' result.

John

 

 

[John Dyson]

We are down to two versions.   I strongly suspect that that one of the versions is correct, and the difference is minute.

 

One version has more expansion, more dynamic, but seems to have a less pretty stereo image.

The other version has less expansion, slightly less dynamic, but a really good stereo image.   Most likely, this second version will be my choice, and most accurate, but there just might be other defects that I hadn't noticed when creating the 'prettier' version.

 

The differences in response are  slight -- both have approx the same sound as to the amount of HF & LF.  The HF and LF are quantatively very similar to the FA original, but with more clear sound.   The difference is mostly about the amount of expansion at different freqs.   Since the darned dispersive expander doesn't provide a natural way to 'measure' the amount of expansion, this is probably the most subjective part of the decoder, which is already too dependent on subjective comparisons.

 

It does seem like the output of the decoder is 'close' to reality, hopefully the right choice will be made -- or if the two choices are incorrect, I"ll catch the problem before the release.

 

John

 

 

[John Dyson]

Sorry for the delay -- both 'good' versions had a slight HF rolloff problem.  There isn't too much HF, and actually doesn't appear to be too weak either -- the potential problem seems almost like a time delay or something odd like that.   Not all previous versions had this characteristic, but also I didn't hear this 'strangeness' until about 1/2Hr ago.  Both prospective versions appear to  manifest 'it'.   I caught it during a spot check of the demos while they are being decoded, and caught it about 2/3s way through the long set of demos.  (Before the ONJ, short demos and a few others.)    I thought that there was going to be an upload a few hours from now, but must be delayed now.

 

Otherwise, the stereo image is very, very good -- almost eerie.   In a casual sense, the sound is nice, seems like low distortion to me. if judging the HF as being too much or too little, I'd guess that it is 'too little' if not a little 'reserved'.   Then, when a lot of HF is called for, the HF isn't missing at all.

 

Given this issue, and the weakness of my hearing, I'll go ahead and let the rest of the demos run to completion, assuming the problem is with 'me' and my hearing.   Then hold off the release until this thing is understood and/or the 'bug' (if any) is fixed.

 

There is absolutely NO, NONE, NADA long EQ other than the descrambler, and it doesn't create this oddity.   This 'sound' can be mitigated by removing the 'final rolloff', but the amount of EQ is very slight.   Something else might be happening, perhaps even an array out of bounds (normally, the code does range checks on arrays, but I might have slipped up.)

 

Gotta hold off the X-version release until the phenomenon is understood.   Should only take a day or two.

 

John

 

 

 

[John Dyson]

Here are some demos as a *status report*, not a release demo.   Relative to this demoed version, there are still two bugs -- both in the descrambler.   The descrambler bugs were just fixed....    The bugs in the version below were strange, STILL IN THE DEMOS, to be updated as the 'release' in a day or so:

 

1)  Certain frequency ranges need to be compressed instead of expanded.   This bug caused some vocal chorus (esp in Elton John's Daniel) to become garbled or too strong.

2)  Most important -- apparently, the descrambler needs to run in mid/side mode instead of left/right.

 

Look for the version V8.1K13R snippets in the directory below, the other demos are also under V8.1K13R...

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

 

*  You *will* notice a weird problem with the stereo image.   The two corrections mentioned before will correct it.   The correction already exists, but any fix like this sometimes creates a chain of additional corrections.   The follow up, true release will likely be delayed for another day or so.

 

This mistake about mid/side is probably the reason for my continual decision problem about using L/R or M/S domain processing.   These two bugs also encouraged a persistent EQ problem, that EQ problem was the original reason for this extra day delay.

 

AGAIN -- these demos are NOT release version, and show that I am not 'holding back' for no reason, and there is true progress.   The demos are misleading because of the two bugs mentioned above.   This L/R & M/S processing matter can happen because of the parametric nonlinearity in the descrambler.   A normal EQ sequence doesn't care what domain is being used.

 

The 'descrambler' is a lot like a 'Russian doll'...   The layers appear to never cease!!!

 

 

 

[John Dyson]

1 hour ago, John Dyson said:

Here are some demos as a *status report*, not a release demo.   Relative to this demoed version, there are still two bugs -- both in the descrambler.   The descrambler bugs were just fixed....    The bugs in the version below were strange, STILL IN THE DEMOS, to be updated as the 'release' in a day or so:

 

1)  Certain frequency ranges need to be compressed instead of expanded.   This bug caused some vocal chorus (esp in Elton John's Daniel) to become garbled or too strong.

2)  Most important -- apparently, the descrambler needs to run in mid/side mode instead of left/right.

 

Look for the version V8.1K13R snippets in the directory below, the other demos are also under V8.1K13R...

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

 

*  You *will* notice a weird problem with the stereo image.   The two corrections mentioned before will correct it.   The correction already exists, but any fix like this sometimes creates a chain of additional corrections.   The follow up, true release will likely be delayed for another day or so.

 

This mistake about mid/side is probably the reason for my continual decision problem about using L/R or M/S domain processing.   These two bugs also encouraged a persistent EQ problem, that EQ problem was the original reason for this extra day delay.

 

AGAIN -- these demos are NOT release version, and show that I am not 'holding back' for no reason, and there is true progress.   The demos are misleading because of the two bugs mentioned above.   This L/R & M/S processing matter can happen because of the parametric nonlinearity in the descrambler.   A normal EQ sequence doesn't care what domain is being used.

 

The 'descrambler' is a lot like a 'Russian doll'...   The layers appear to never cease!!!

 

 

 

Just wanted to forewarn -- the eventual release sounds a LOT better than the early example above.   It is amazing how quickly the sound can improve with minor changes, esp in the descrambler.

I strongly suggest that the resulting sound is/will be impressive, yet reserved.

 

John

 

[John Dyson]

Still working on the release  -- really trying for 'as good as possible' -- has a notably 3d kind of sound now.   I have been learning exactly how to control the descrambler.  The structure of the descrambler is not as much of a challenge as how it was used.   A properly set-up descrambler is paramount for the best reproduction.

 

(If this message seems to be more rambling than usual -- sorry about that, but my 'brain' is focused on the decoder...   I wanted to mention that progress is continuing, and that I have waken up to the numerous things that I have overlooked in the design!!!)

 

In the descrambler, there are some 'swaps' between compression and expansion, which I had previously missed.   The all critical 4.5kHz region was previously totally botched, and the descrambling in the 1.5kHz to 3kHz was also  previously botched.   The older versions of the descrambler would magnify the sound of vocal chorus type sounds, especially male/female duets, where it might sound like 'more detail', but instead was 'more distortion' of sorts.

 

In a way, the 'magnification' of the interactions in vocal chrous would make 'Take me Home, Country Roads' from ONJ sound like there is more detail.   That 'greater detail' was purely synthesized by the mix of compression and expansion.  It also made 'Supertrouper' from ABBA sound somewhat garbled.

 

This longer than expected development delay has resulted from both  iteratively searching for the correct descrambler settings, then waiting 3-4 Hrs for my hearing to 'rest' so that I could be successful judging the sound of the descrambler output.   This is a back-and-forth test cycle, in all cases the goal is the accurate response....   'Sounds good' is less of a criteria than 'is likely accurate.'

 

I could release something NOW, but as you might have noted in the 'early demo' a few days ago, the result wasn't optimal, and still needed work.

As soon as a release candidate is good-- I'll do the release....

 

No more 'demos' until the actual releasable code is used.   Today is Wednesday, and I'd like to make the release before this weekend.   The release can not be rushed.

Private requests for demos are welcome, and at this point the sound is good, but still need to make sure the HF rolloff is correct (my hearing deficit causes troubles, and stuck with 'sounds good', which is not good.)

 

John

 

[John Dyson]

It is likely that there will be an experimental (not preliminary) release in the middle of this weekend.

There has been major improvements in the descrambler* as my understanding evolves.  There is a lot of 'back and forth' evolution, with EQ blocks being added and removed as the result improves.   There have been a few cases where the work needed to start from scratch with an alternative choice being made early in the process.   The technical part, understanding the basic processing is trivial, but second guessing the original engineer (engineering team) has been tedious at best.

 

It is time to do a release.   When reviewing the results of the demos, it has been very difficult to quit listening.  The recordings sound SO GOOD that it is hard to quit listening.   The major defect in the upcoming release will probably be some left over garble.   Any little error can hurt , but the errors are slowly being decreased.   The current acitivity is the 'back and forth' attempt to find the correct settings to zero the errors (diminish the left-over garble.)

 

*  I use the term 'descrambler', but a more technical description might be 'dispersive compressor/expander'.   Instead of 'one' dispersive processor, it is actually multi-band and really scrambles the signal, sometimes in suprising ways.   Once descrambled, the recording can sound significantly more clear.

 

Even IF this upcoming release isn't perfect, even though it might be somewhat surprising, it might be interesting enough to give it a try.   I won't produce any easy-access demos until the full array of private and public demos are uploaded.   The only reason for the careful staging is that I just might have to rescind the release.   Unfortunately, I do make mistakes resulting from too much enthusiams, so a last minute pull-back is sometimes needed.   After the announcement, I'll make a posting a few hours afterwards to verify that the uploaded release is usable and ready for casual review.

 

If the upcoming release isn't rescinded in an hour or two, it will be available until the next update, hopefully a true V8.1K instead of 'X8.1K'.

 

The descrambler was originally a last minute surprise, then the need for different settings at each frequencies was another 1-2month surprise.   Now, this rather major change in the EQ & pre/de-emphasis & mid-side conversion was another surprise just recognized over the last week or so.

 

There was going to be a release two weeks ago, but the descrambler has been mind blowing...   I regret the delays and the historical/hysterical mistakes, but this delay is very much worthwhile.

 

There *WILL* be a release in <+27Hrs unless there is a major disaster (personal or otherwise.)  In the worst case, the current test version could be uploaded -- the current results are definitely better than any previous version and/or the original FA version.

 

John

 

 

 

[John Dyson]

As promised -- there is now a V8.1K-X8.1K-100 available.   Unfortunately, it is probably not of release quality simply because I cannot judge through the last day or so.   There might be an HF/MF/LF balance problem that results from a simple descrambler setting.  (Ran out of my hearing medicine, just got a new scrip, but takes a day or so to recover.)    Once I know which setting is correct, the change to 'correctness' is truly almost trivial.  (Frustratingly, I lose both HF and LF individually, so the results are insanely difficult to choose.)  

 

Please tell me:

1)  Just right

2)  Too heavy

3)  Too much HF...   (unlikely.)

 

 

The decoder is not yet uploaded, but the snippets are available at the location below, and other stuff is available in the regular places.   Once I get some feedback as 'too heavy, needs midrange/HF' or 'just right', then the 'full release' will be made available within a few hours.

 

THANKS -- and if this is the correct choice -- PLEASE ENJOY.   If the different choice is needed, then please enjoy once the decoder setting is changed....

(It is SOOOO EASY to change the setting, I just cannot choose.)

 

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

 

 

[John Dyson]

4 hours ago, John Dyson said:

As promised -- there is now a V8.1K-X8.1K-100 available.   Unfortunately, it is probably not of release quality simply because I cannot judge through the last day or so.   There might be an HF/MF/LF balance problem that results from a simple descrambler setting.  (Ran out of my hearing medicine, just got a new scrip, but takes a day or so to recover.)    Once I know which setting is correct, the change to 'correctness' is truly almost trivial.  (Frustratingly, I lose both HF and LF individually, so the results are insanely difficult to choose.)  

 

Please tell me:

1)  Just right

2)  Too heavy

3)  Too much HF...   (unlikely.)

 

 

The decoder is not yet uploaded, but the snippets are available at the location below, and other stuff is available in the regular places.   Once I get some feedback as 'too heavy, needs midrange/HF' or 'just right', then the 'full release' will be made available within a few hours.

 

THANKS -- and if this is the correct choice -- PLEASE ENJOY.   If the different choice is needed, then please enjoy once the decoder setting is changed....

(It is SOOOO EASY to change the setting, I just cannot choose.)

 

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

 

 

I have decided to produce the 'less heavy' version also.   Both are similar -- when I hear well, I seem to prefer the first version.   However, since I lose LF as well as HF at different times, this is a bit of trouble.   There are a few settings that produce consistent results.    When the results aren't consistent, the stereo image flops all over the place.   Even though I say 'too heavy' in the description, it isn't quite the same thing as too much LF, even though it seems similar.

 

It will take much of the day tomorrow (or today, it is 0020 my time.)   Family activities today.   After the family activities, I'll change the settings and build the different version.   Both versions are consistent, and there should only be a slight difference.

 

John

 

PS: the design is NOT based on freq response, and when running the alternative version, there is little difference in 'freq response', but the dynamics are different, perhaps improved in some ways.   The sound seems 'more smooth', but also a little more bright.   It might be more correct, but cannot be ready until at least +15Hrs, probably +24Hrs.

 

The sound DOES seem more 'pretty' and more 'clean' in some ways.

 

 

 

 

[John Dyson]

Well -- when making the minor change for the slightly different choice, attempting correctness, I found an incorrect choice for one of the EQs and an incorrect descrambler setting.

The new update will have the more correct decoding parameter, but will also have the alternative descrambler settings.

 

The 'effective' EQ is slightly different, along with a slightly corrected LF (it wasn't as flat as it should have been).   The LF will be about -0.75dB in the 100Hz +-50Hz range relative to the previous version.   The change makes the bass in the 100Hz range about +0.1dB too strong.  The width of this +0.1dB error is in a similar range of 50Hz to 150Hz.   This +0.75dB error mistake leaked through, and the LF EQ is not a 'conventional' kind of simple 1 pole and/or 2 pole EQ.   Instead, the EQ results from a combination based on the 2.213 ratio and a 1.25 component.  It is now easy to automatically compute these frequencies, but  the need for specific settings was still a designers choice.

 

The LF & MF can be measured, and are working as expected.   WIth the descrambler, the EQ from 1.5kHz to 20kHz+  has a native +-1.5dB (at least) measurement error.   The HF EQ is certainly within that range, but since the descrambler has a wildly variable gain, not based on any time constants, it is impossible to measure without a reference standard.  It has been very fortunate to have found a moderately highly compressed commercial recording with wide bandwidth, which gives a good source for generating an estimated frequency response measurement in the HF.

 

HOWEVER -- the HF EQ has been close to correct in the last few weeks, but it is the *expansion* that now has been a bit of trouble.   Without descrambler expansion, the sound ends up being dead and lifeless, obviously incorrect.   It is very easy for the descrambler to expand sections of the signal by 3dB or more, with the result being unbearable.   Unfortunately, the descrambler is an integral part of the HF EQ, therefore it is tricky to get a reasonable frequency response measure.

 

After a LOT of trial and error, the general EQ does appear to be correct, but it is very likely that the PREVIOUS VERSION made available today (-100) descrambler setting was too restrained.   It is a tricky thing to decide the correct combination of parameters.  The new version, X8.1K-103 has the 'less restrictive' settings, more in the center of the range of likely correct settings.   The version -103, demos being run right now, will also be accompanied by the decoder itself as V8.1K-X8.1K-103, hopefully the preliminary for V8.1K.

 

John