EC Designs

[loandsound]

(02-27-2018, 12:29 PM)paulw a écrit : Une fois le "super algorithme" de contrôle de volume digital de  leedh disponible dans Audirvana Windows peut être que je ferai l'essai sans preampli en comparaison avec d'autres DAC.

Je me demande d'ailleur si le problème est de trouver le bon algorythme (non pas que cela me semble aisé), ou si c'est d'abaisser le bruit en sortie du lecteur ou (mini)PC, qui prend des proportions d'autant plus importantes que l'on abaisse le volume depuis le lecteur...?

[a supprimer merci]

Une petite mise à jour sur ce DAC (Mosaic UV).

D'une part, Hifi-Advice, qui avait fait parler de la première itération du DAC (Mosaic T) n'a toujours pas obtenu un "review sample" de la deuxième itération du DAC (Mosaic UV), mais fait part des commentaires suivants sur son site (en fevrier):

"With the EC guys, one can never tell… They were going to send a review sample after the next modification that they were working on. Most likely they have again come across new developments that have triggered new ideas for the next product iteration and they are experimenting with this right now.

All I can say at this point is that the UV is very, very good. Much better than the original Mosaic. A mutual friend has one that he brings along every now and then. We have compared it to some very pricey DACs. It has always either held its own or actually beaten the more expensive product."

Effectivement, les frères Brown (ils sont donc deux), toujours à la recherche d'améliorations, sont en train de revoir le Mosaic UV pour:
- intégrer une alimentation sur batterie
- revoir les circuits du DAC afin de les rendre "insensibles" au signal USB en entrée.

Quelques explications obtenues par mail (en avril) sur ce dernier point:

"All DACs are sensitive to the USB audio source, even very expensive (Eur 30,000 and up) DACs from established brands. This occurs with verified bit-perfect playback so the only remaining cause for this has to be the ripple and jitter on the USB signal and the noise on the 5V USB bus voltage. The USB noise spectrum is the sum of all noise sources within the source and being fed into the source.

However, the 5V USB bus voltage is one of the causes that can be fixed easily.
the USB data also contains jitter and ripple that enters the USB audio receiver and all connected circuits because certain bandwidth is needed to get this data through. Circuits within the USB audio receiver add more harmonics (fast digital logic circuits).

Fast logic produces harmonics that extend far into the GHz range, these can spread very easily through stray capacitances throughout the circuits and wirelessly (EMI).

The most obvious cause for degraded sound is master clock degrading (increased jitter and ripple) as a result of the leaking USB data noise spectrum.

Then there is D/A converter circuit pollution through noise (increased noise floor, masking).

Everything connected to the USB audio source also has impact on the jitter and ripple on the USB signal, including WIFI and Bluetooth enabled on the source and hard wired connections to the source like LAN for example.

The best solution would be attempting to reduce DAC source dependency by making both master clock and D/A converter immune to this USB noise spectrum. This is what we tried to achieve with the new Mosaic II DAC.

And it seems we finally succeeded by placing the D/A converter in quarantine (electronic air lock circuit) and developing a novel sine wave clock with nested bandpass filtering (pure sine wave output) and multiple, isolated transformer coupled outputs with very low output impedance of less than 10 Ohms (transformation ratio). Load fluctuations are sensed through the output transformer and automatically compensate the oscillator frequency. This way the output frequency can no longer be influenced by the connected polluted clock loads."

Je n'ai pas de nouvelles depuis, mais j'ai hâte de pouvoir tester ce Mosaic II qui j'espère sera vendu à un prix aussi raisonable que ses prédécesseurs.

A suivre...

[0000]

Leur démarche me paraît la bonne. C’est aussi mon opinion depuis toujours.
Il n’y a pas besoin d’appliquer les recettes audiophiles sur la partie informatique. Il suffit une bonne interface usb dans le DAC.

[a supprimer merci]

Oui, en 2013 John Swenson ecrivait ceci (ici https://www.audioasylum.com/cgi/t.mpl?f=...o&m=125989), qui à mon avis explique clairement le problème. La solution n'est pas encore trouvée !

"I'll try and give some real life examples here of how things other than "the bits" could have an affect on the analog output.

I've been working on a new USB DAC design recently, so I have a setup that I'm continuously looking at with scopes and logic analyzers etc. In this situation the logic analyzer said everything was fine, the bits were perfect. A logic analyzer runs the analog voltage on the wire through a "threshold" to distinguish if it is high or low, what you see on the screen is just high or low, ie "bits". But when I looked with the scope which shows the actual voltage levels of the signal I saw some extra signal riding on top of the highs and lows of the "bits". This turned out to be noise on the ground plane caused by the processor that was generating the bits. (it was much worse than it should have been due to a poor board layout of the processor reference board) That noise was enough to cause significant change to the audio out even though the "bits" were correct.

One interesting aspect of this was that you could easily see changes in this ground plane noise depending on what the processor was doing. While this was a fairly gross example of the effect, it clealy shows that things going on in the processing and transmission of the bits can have an affect on the sound at the output, even though the correct bits get to the DAC chip.

Next you might ask "well isn't that a broken system, if it was "good" shouldn't it not be an issue?" Note that this was the official reference board for the processor made by the manufacturer, who should know how to make things that work well with their processor. This just goes to show that things that can cause audible differences in digital audio are frequently not part of "it works as a digital system", the board did what it was supposed to, it delivered the bits.

A better board design could have cut this ground noise down significantly, but it would still be there.

What we DAC designers have to do is figure out ways to design products that produce analog out that is immune to this sort of thing. Unfortunately this is extremely difficult to do. There are many people on this board that expect that this is easy to do, just put in the right 50cent part and presto the design is completely immune to everything. It doesn't work that way. High frequency ground noise is extremly pernicious stuff, it will find a way to get around just about any obsticle you throw in its path.

Different designers take different approaches to try and achieve this with varying degrees of success. The different approaches will usually be affective at decreasing susceptibility to different types of noise so one DAC model may not care about a certain aspect (say cable differences) while another may be pretty immune to cable aspects but be susceptible to timing variations in packets. This may be a part of why some people say they can hear certain aspects and others say they cannot.

These techniques for noise suppression are pretty esoteric knowledge, there really are only a few people that really understand all this, there are very few places in the real world were the combined knowlege to make this really work right are required, thus very few people have a good grasp on all of this. The result is that many actual designs on the market are fairly lacking in this department, or are only targeting one aspect of it.

This is slowly changing and companies are starting to get an inkling of what it takes to do well with this and are hiring people with some knowledge in this field, but there aren't nearly enough to go around, so it's going to be some time before all digital audio systems you can buy do a good job in this regard."

En plus du bruit généré en amont, il y a également du bruit généré dans le DAC lors du traitement du signal USB en entrée, et dans les autres processeurs utilisés...

[a supprimer merci]

Le nouveau DAC progresse. Voici quelques informations:

"It is a S/PDIF Toslink DAC based on the DIR9001. It comes in 17 or 24 bit version and it is battery powered (3V7 Lithium Polymer battery).

The battery can be charged with a standard 5V adapter through the USB charger socket on the top right.

It has a standard rectangular Toslink optical socket (next to the USB charger socket).

Mechanical switch connects the battery to either the charger or the circuit (both plus and minus are switched). So the DAC circuit is never connected to the charger when switched on.

The 10Ah battery (on the picture at the right) can power this DAC for 285 hours.

The DIR9001 has a modified PLL for low very low jitter.

Hardware decoder extracts sample rate from the DIR9001 and drives a dual white seven segments display (-- not locked, 44, 48, 88, or 96). There is a photodiode that automatically controls the brightness.

Second hardware decoder converts 24 bit left justified from the DIR9001 to either 17 bit or 24 bit data & clock burst signals the drive the Mosaic converters. 

So there is no micro controller and no master clock in this DAC, therefore noise levels and EMI are extremely low and power consumption is only 130 milli watts during streaming so the battery lasts very long.

These are new D/A converters (routing was quite a challenge). These are based on the mirror bridge concept that offers superb low level performance, even with 1% resistors (I use 0.1%) and is rather insensitive to jitter as all interface (clock and data) signals are completely shut shut down well before latching the outputs. 

There is no CPU nor crystal oscillator running that could cause interference. The DIR9001 consumes only 26 milli watts and produces very little interference. I put it as far away from the converters as possible to ensure noise levels stay low.

Huge decoupling caps (I plan to use 3 x 10000uF ensure that the battery power supply approaches a perfect power supply

This way the digital interface noise level is almost zero (no jitter is transferred) when the latch pulse occurs.

This requires very fast logic (CPU is not fast enough) as all data has to be clocked into the converters in the fraction of the time that it would take to clock in I2S for example.

The output (4Vpp, 750 Ohm) are directly connected to two RCA sockets on the top left.

And that's about it, a very simple, straight-forward DAC that outperforms all USB versions we designed and tested."

[a supprimer merci]

La production du nouveau DAC est en cours. Il y aura deux modèles: 16 bit et 24 bits. Les prix sont très raisonnables (moins de 1000€). Voici une photo.




Pour mémoire, il s'agit d'un DAC avec entrée Toslink uniquement, alimentation sur batterie (très faible consommation), et utilisation de deux TDA1541A. La batterie de 10Ah fera fonctionner le DAC environ 600 heures. Le port USB en haut à gauche est utilisé pour recharger la batterie. Des techniques innovantes (voir précédents posts)  permettent de réduire considérablement le bruit au sein du DAC. L'utilisation du Toslink, avec un reclocking dans le DAC, permet d'obtenir une immunité à la source (c'est ce qui est affirmé en tout cas). Si c'est le cas, et je n'ai pas de raison de douter leurs affirmations, compte tenue de la qualité de leur précédent DAC et des mesures qu'ils on produit, ce sera une petite révolution dans le monde de la demat ! Exit les streamers à plusieurs milliers d'euros et les solutions complexes à base de PC optimisés...

Le petit module présenté en bas sur la photo est un adaptateur USB - Toslink. Quelques mots sur ce module: 

"This is a tiny (USB stick sized) USB to Toslink converter. It plugs directly into the computer, eliminating USB interlink issues. It has a low jitter master clock and a nano power CPU (33mW) for lowest obtainable noise levels. It has an on-board ultra low noise voltage regulator and all USB pins are protected by diodes..

It has a dedicated Toslink transmitter chip for low jitter Toslink output.  Serial output can be used to drive external volume control units through PC, pad or smartphone. Because of the design we can offer optimal performance for a low cost. The USB audio source no longer seems critical at all as long as it's bit-perfect."

Trois autres produits seront disponibles: un contrôleur de volume, un lecteur de fichiers sur clé USB à base de processeur RISC à très faible bruit, avec sortie I2S et Toslink, et un amplificateur 2x25w.

Concernant le contrôleur de volume: "The professional circuit boards for the SVC (Servo volume control) also arrived. This volume control is based on two motorised Bourns high endurance carbon slider potentiometers (studio quality) controlled by a small micro controller. It is battery powered (single 3V7 Lithium Polymer battery). Prototype is already up and running, it was quite a challenge to design a digital servo system that uses very little power and can adjust the sliders to 0.1% accuracy. The motors move the slider from min to max within one second and the motor won't run at all with voltages below 5V, so these are the most difficult to drive servo systems imaginable. I use a DC-DC converter to get enough voltage for the motors from the 3V7 Lithium battery. Microcontroller and DC-DC converter switch-off after volume adjustment so there is zero interference after adjusting. The servo first performs fast course adjustment followed by micro-step fine adjustment. Volume control is completely pop and click free and the muting works with smooth fade in and fade out."

Concernant le lecteur de fichiers: "a purist battery powered WAV USB stick player (44.1/16 ... 96/16). It supports FAT and EXFAT, up to 99 CDs with each 99 tracks on one stick. The UPL can support highest capacity USB sticks available. It also has shuffle mode and comes with an innovative IR remote control based on two square control surfaces. each 4 x 4 cm control surface has up to 8 functions.This USB WAV player runs on a single RISC processor from ST (it produces far less interference than any PC or professional (PC) based streamer, it produces also far less interference than a RPI. Estimated power consumption is roughly 370 milli watts during playback. because we use ultra fast and silent DMA (Direct Memory Access) that only supports 16 bit blocks we cannot support 24 bits at the moment. Adding 24 bit support would require (partial) DMA bypass and this would require a far more powerful CPU with hardware USB to parallel converter (PHY chip) that could no longer be battery powered and would produce much more interference.

The UPL has both, Toslink and I2S outputs D, BCK, WS, MCK (RJ45 connector), so it is also interesting for driving existing DACs"

Et les amplis:
"It is based on a novel double delta input stage and has circuits that eliminate even smallest ripple and noise voltages (cascaded capacitance multipliers on the power rails and double low pass filtered bias reference voltage for the lateral MOFETs). Other unique feature is variable gain (5x, 10x, 20x) using a front panel 3 position switch and a display. This enables lower attenuation settings (by using lower gain setting) resulting very high resolution and extreme SN ratios that are not possible with existing amps.

Bandwidth is very high, 2 MHz @ 5x gain, 1Mhz @ 10x gain and 500 Khz @ 20x gain while maintaining excellent stability, this is possible with the double delta input stage. The large bandwidth ensures absolute minimum phase errors and a very realistic sound reproduction. The MBL is based on global feedback for very low distortion and very low output impedance, global feedback now works fine because of the ultra fast novel double delta input stage.

The MBL idles at approx. 12 watts (200mA bias) and outputs up to 25 watts in 4 / 8 Ohms. Both input and output are DC-coupled for optimal transparency"

[volpone75]

(09-03-2018, 11:02 AM)paulw a écrit : .... Si c'est le cas, et je n'ai pas de raison de douter leurs affirmations, compte tenue de la qualité de leur précédent DAC et des mesures qu'ils on produit, ce sera une petite révolution dans le monde de la demat ! Exit les streamers à plusieurs milliers d'euros et les solutions complexes à base de PC optimisés...
...

Interessant. Merci pour le suivi EC Design.
En terme d'argumentaire technique c'est convaincant en ce qui concerne les moyens employés pour l'isolation et la résilience au bruit du convertisseur. A confirmer "en vrai", à l'écoute. Quelle est la résolution et la profondeur de bit max supportée en entrée ?

[a supprimer merci]

24bit et 96khz.

[Jacques92]

(05-22-2018, 11:46 PM)paulw a écrit : And it seems we finally succeeded by placing the D/A converter in quarantine (electronic air lock circuit) ..."

Ah bah zut c'est exactement ce que je suis en train de développer dans mon petit coin depuis quelques semaines.

Pour la nouveauté, je crois bien que c'est raté 

[a supprimer merci]

Voici les principes:

" there is a PLL master clock in this DAC, it is located inside the DIR9001 Toslink receiver. This chip already offers -very- low jitter (max. 50ps with the datasheet loop filter).

I modified its PLL loop filter so jitter stays even lower (even with very high jitter on the Toslink interface). It now acts as a flywheel, it can only -slowly- follow the input frequency changes, too slow to pass on jitter. The only disadvantage is that it takes longer to lock (approx. 1 second instead of milli seconds).

Crystal master clocks emit a powerful spectrum that reaches 10 meters and more, so it is very difficult to shield when its located inside the DAC right next to the D/A converters.

Low phase noise (jitter) crystal oscillators emit one huge peak signal at resonance (fundamental frequency), for example 11.2896 MHz, this is problematic. The master clock inside the DIR9001 is not very powerful (the complete chip draws only 25% of the power a single crystal master clock would typically consume) and has a spread spectrum that spreads this peak signal for reducing EMI peaking.

I only use low frequency clock signals from the DIR9001 (Highest equals 6.144 MHz at 96 KHz sample rate and only 2.8224 Mhz at 44 KHz sample rate). This is easier to handle compared to say a 48 or 96 MHz crystal oscillator plus high frequency audio clocks of 22.5792 and 24,576 MHz used in conventional USB DACs that produce a very powerful, highly complex summed noise spectrum.

I developed a new hardware decoder that exposes the DAC to the jitter spectrum (noise) for only 16 or 24 of the total 64 bits in a frame. This reduces the impact of jitter by 75% (when using 44.1) compared to I2S.

The novel Mosaic Mirror DAC is no longer sensitive to jitter like conventional DACs are. It consists of two identical but mirrored D/A converters that are merged together. One outputs a signal from +4V and down while the other outputs a signal from 0V up at the same time, hence the name mirror DAC. The signals meet at the centre (0V) where the signals and the noise on these signals fully cancel.

So I basically generate the output signal by varying the degree of signal cancellation. This way noise around the critical zero crossing area is lowest as both signals plus the common mode noise on it will cancel.

For clarity, jitter does change timing ever so slightly, this however is completely inaudible. Few pico seconds sample timing change is inaudible. The noise jitter dumps on the DAC output through demodulation is very audible, especially around the signal zero crossing area when jitter is no longer masked by louder signals.

The jitter riding on clock and data signals that reaches the DAC output will cause audible changes that are related to jitter. This is where I tackle jitter, I prevent this noise from reaching the DAC outputs by simply shutting these signals down completely for as long as possible.

I attached an oscillogram illustrating what I mean.

You see a 16 pulses (data & clock being injected into the DAC), followed by 16 bit silence (interface shut down completely). Then there is a small spike (sample is latched to the outputs) followed by 32 bit silence (interface completely shut down again).

I2S would switch all of the time (it never shuts down) so we get 100% jitter exposure (the jitter is IN the data and clock signals).

With the Mosaic Mirror DAC we only have 25% exposure as you can see on the oscillogram."



C'est également décrit sur le fil DIYAudio à partir d'ici: http://www.diyaudio.com/forums/digital-l...56357.html

Les commentaires suivants dans le fil sont intéressants. 

P.S. Ne pas me demander d'explications ! Je ne fais que rapporter...