Dear all,
A few -I think- important milestones have been achieved, so I thought to inform you about it.
Hopefully it also shows better why things seem to take a long time, while it's really worthwhile the long waiting. And, I'd hate to pronounce improvements later, while some of you may have a "less" version in the mean time. So it better be right the first time !
Allright. Starting with what I mentioned in my before post, production really has started a few weeks back. A 100 units will be produced as a first batch (I may sell 10 only haha), so hopefully enough people can try the DAC (money back guarantee) without being in a long back log while others already have it. Thus, only when the 100 are ready, the first will ship.
In the mean time, this will give me the "opportunity" to wait for some parts with a long lead time. This just is so and can't be helped, but for some parts only the best can be used, and they are produced on (my) order. Currently the longest pending lead time is 8 weeks.
But, these are parts which can be assembled quickly, and the assembly won't hold up production really (the waiting for the parts does).
Although you didn't know it, only yesterday another technical highstand has been achieved, and I must say this is about one of the most nice (and difficult !) coorperations not in the same office I have ever met in my life. Think of hundreds of emails about a certain design at the bit level (this is literal), that in the end needing independend programming at two sides, one side being XXHighEnd outputting data, the other side being CPLD programming, which is actually in-DAC and at the engineering level. And, after 6 months of careful preparation, yesterday it worked in one go (neither side being able to test without the other's efforts).
Some know (via offline) what this is about, and actually it is about creating some kind of "double buffer" to be 100% sure it can't be done better. It is about the 24/384 input of which we can reason that better (higher sample rate) isn't necessarym but because the technology to go further was there in (my) theory in the first place, it just was done : 24/768. Mind you, this is for input, and no master files exist for it.
However, since Arc Prediction Upsampling turned out to be a quality phenomenon by itself, it *will* be useful for just that : sound quality.
On this matter, please notice that the step from 24/176.4 to 24/352.8 (with a source of plain 16/44.1) already brought a significant improvement, so stepping further to 24/705.6 most probably will again (I couldn't test this myself, because I don't have the hardware for it yet).
By pure coincidence, also yesterday I was able AT LAST to apply a gain "mechanism" with THD+N specs better than the DAC chips itself, or in other words, which does not degrade sound. Maybe, just maybe I have to come back on this later (because I don't like the sound of it afterall), but chances are very small because of the enormous improvement ot brought (see below story).
This must have been the most time consuming of it all, and in the end is about a decent I/V stage which most probably is the most discussed subject in the DIY community on the Internet. This is also about my earlier expression on "you have to live with 24dBFS under 2VRMS" (which by itself is very doable with enough gain in the main amps and/or enough efficiency in the loudspeakers), which now is just 2VRMS (which is the standard). Also, this will allow for headphone output, although in theory only (read : this has not been tested yet).
To my own surprise, the at last working of this, brought a new dimension in music reproduction through loudspeakers for me. If I had to rate the change in SQ, I would say "10 times better". Again ? yes, again. Look :
People who are here from the beginning (or those who ran into the post(s) concerned by accident) know that I have always had one major negative remark about music reproduction through loudspeakers : the volume level of cymbals and the like. And since we are talking about it anyway, you may come to the same conclusion while I put your attention to it :
Go to a live perfromance, and have the bass player, the singer, the piano player and the drummer (etc.) in front of you. Now put your attention to the balance of all the sounds *but* the cymbals on one hand, and those cymbals at the other. What actually sounds the most loud ? It is the cymbals !
Now go back to your listening room, and try to perceive the same balance. No way it is going to be met. Ha ! you are even lucky if you perceive (real) cymbals in the first place !! ... so bad it is ...
On this matter, maybe three years ago by now, I tweaked my crossovers so I would have 16dB more output at 20KHz, with a nice slope starting at 5KHz and up (to that max at 20KHz). That really helped a lot, and was actually the first time hi-hats became nicely profound while before I never heard them as an explicit instrument playing (while it is the most important instrument for a drummer).
I never removed that tweak up till today (and I still won't). So, although this may sound wrongish to you, it is my opinion this is just needed in order to receive a more or less representative playback for cymbal like instruments. Also, this can't be blamed to our ears, because I really don't have a problem with the drum kit in the house here. The live cymbals perform perfectly right and they really don't need cranking up.
So what is going on really ? And to remember, a self respecting loudspeaker just measures flat.
I think I now found the answer in a (to me) new phenomenon - or dimension if you like : The SPL of high frequencies.
Oh yes ...
I am quite sure (not 100%) that most of you look at SPL (Sound Pressure Level) as a phenomenon which occurs in the lower regions (I mean under 5KHz or so). SPL is equallish to sound pressure you can feel, and it comes from base drums, general low frequency sounds and maybe ar loud singing voice. But not from something like a tambourine. No ?
Ha, come here and be startled. And this is exactly why I rather call this a new dimension than just a phenomenon. It makes your whatever music completely new. There's a whole other band playing besides the one you were used to. It is the drummer with his gags.
Btw, I must admit, the mid level increased with it, and those sensitive to that will be quite happy just because of that alone. But that's another subject (as important btw, but not as a new dimension as such).
Quite a few virtues spring from this, like really fresh sound is one of them. So, people who tend to talk in terms of removed blankets ... here you go ! it is really amazing.
But what actually happened ?
To be honest, at this moment I don't know yet, although some of it could be predicted from earlier attempts in a similar setup. I think the other day I talked about cymbals sounding for 20 seconds instead of the "usual" 5 seconds, which I remembered from this earlier setup, though not good sounding. So, it seems logic that (relative) additional amplification of the high frequencies make jump out a cymbal for 20 seconds long, while before it could only jump out for 5 seconds. It was there for 20 seconds allright, but the sound of the piano etc. made it disappear. Not so anymore. And, obviously along with it goes the as louder attack. So, just louder (and not a bit, but severely more).
If I had to explain it from a technical point of view, I would say this :
The setup from before was about speed. You have read about it, just explicit speed. But since this is a.o. about leaving out parts, it is also about leaving out "drive" (notice : in my case, and of what I can reason only after the happening). Now, we tend to derive drive from low frequency sounds. In other words, I think it is recognizeable that when we don't have enough drive, the sound gets too thinny and the bass disappears. And, since this was not the case in my setup, I had enough drive. However ...
A low(er) frequency sound moves slowly for its up and down (plus and minus voltage) wave. It is not difficult for such a wave to go down again after it reached the (volume) top, because it is slooooow. It needs power (yes, drive) to go up and down, and it even needs additional power when the frequency is low because it has to move a lot of air (simply said, and expressed in the wrong domain perhaps), but what about the high frequencies which have to be pulled back in time ? So, a high frequency may not need much power to move the air (which is related to the directionality of it !) but it still needs power to retract from its top in the wave back down to its low side, etc. And *this* goes 1000 times faster for 20KHz opposed to 20Hz. It may even need more drive to control it, and in the end is similar to reracting a loudspeaker diaphragm from its excursed state back to the other side. If this would be left to the mechanical properties of the diaphragm it would be way too slow ...
Further, try to imagine that everything which is too slow to cause that reraction, will cause peaks not to be reached (because the power to create the peak is slow, the creation of the peak is behind, and before it's at it's top it has to go down again ... and peak = SPL (volume)).
The above summarized :
While I had created a most speedy setup with lowest distortion figures possible, apparantly (!) this setup didn't contain the drive to perform in the high frequencies. Notice also that without a speedy setup, the higher frequencies don't perform for the same reason (all is too slow again, read above alinea). Now though, I was able to create a situation with the same speed, but with the power needed. Ehm, apparently, and as long as my own reasoning will stand.
Still here ? well, get intrigued on the next then :
I have always been shouting that my cymbals improved and improved, and whereever I go they are not there at all. But, whenever I said that, I also said (a bit depending on the stage I was at) that they could be improved still. One time they are too plastic, the other time they hiss too much, and the next time the color ain't exactly right. But please imagine, this always was in the "environment" of the cymbals being too much in the background. And as I said in the beginning, many of you will be glad to perceive them explicitly in the first place, although you always will be hearing that they are there. Also, even with my 16dB increase of the higher frequencies, it was allright for me, because they were not too profound (as said, still in the background). But hey, not so anymore !
And so you can imagine that I created myself a huge problem, because at any smashing cymbal in your face, they better sound real !!
Thus, in the very end I may be much excited what has been achieved on this, and it may sound the most interesting to me at the moment, but the next thing now is to get it completely right. It needs much more albums to play to perceive the real merits of what I just said, but I expect some problems here. One thing : I already know that the cymbals sound way better than before for their colour, so nothing turns out to be worse opposed to before. This means that the route itself is the right one which is important for the need to (not) undo that. It is to be awaitened though whether I can "hear" where to tweak further, but hey, I have some 8 more weeks for that (according to the beginning of this post), because this too is back end stuff.
Last thing for now :
Currently I am working on the rather tough decision to lower the jitter which officially is under the audible level already (4ps RMS max);
The reason to do this is an attempt to remove the influence from software (players). This anticipates on software influencing the jitter level itself, which by that means reaches the audible level (like 100ns or whatever). Notice that this possibly happening is only a thought and can't be proven (measurement will show just 4ps, but measurement implies a fluent load on current and is no music (with varying load)). The idea is that implementing a much lower jitter clock (if possible at all) will let rise the jitter again during music playback, but not above audible levels.
So, *or* I may succeed in finding a lower jitter clock solution and if I'm right we can throw out Q1 with unpredicted result (will SQ now always be at its best ??) *or* I let it be deliberately as it is now, so we can keep on using XXHighEnd as we are used to (and have fun with !).
The latter seems stupid to do when the first would be feasable. This is why it is a tough decision (and this too is a back end application).
Notice that with the current (4ps for net measureable result) XXHighEnd influences as much as ever before.
So far for now !
Peter
