Asynchronous I2S FIFO project, an ultimate weapon to fight the jitter

J, I wouldn't call the audiocircle link I gave above "an unsubstantiated rant". There is plenty of justification and measurements to support Pat's claims. There is even more in other threads but I don't have the links and cba to look them up.
I guess you are referring to the link abraxalito posted, but that's only Jocko venting off! Always fun! :D

IMHO, this is a very interesting project and it'd be nice to be improved further (by comparing/measuring clocks or something) than getting stuck on which transformer to use.
 
J, I wouldn't call the audiocircle link I gave above "an unsubstantiated rant". There is plenty of justification and measurements to support Pat's claims. There is even more in other threads but I don't have the links and cba to look them up.
I guess you are referring to the link abraxalito posted, but that's only Jocko venting off! Always fun! :D

hehe yeah i was referring to the Jocko rant, all good i know what hes capable of, so i'm only poking fun. i'm not sure if hes done any testing in the last couple of years as i believe they have added some rather wide band parts, possibly as a result of his criticism. initially i wasnt impressed with the spec either, taking into account the price especially, but the ones mentioned seemed to me to be fine in that regard.

its kinda moot at least for me, as i mentioned its not a high priority of mine anyway and dont want to get into the fine details of transformers when as you said, there are much more interesting things about the project. i would however enjoy playing with these other functions it has for both spdif and i2s buffering and clock switching/distribution.


and no i'm not getting into that with you abraxilito, i'm beta testing a device that seems to cover all bases, but cant really give out detail. i've been accused by some members wanting to distract from their own issues of promoting it for personal gain. but ive been pretty upfront about the fact that i have it for testing since i was approached to help with feedback on the unit, so that must be taken into account by anyone who reads my posts on the subject when i post with more detailed impressions of it's operation.

i'm yet to test its multichannel abilities till i'm finished my 4 diff channel quad mono power amps (w/matching 4 diff sabre channels, not dual mono, dual stereo diff). i've only had it running 2ch at home at all for about a month as i only recently got a new mac mini that can run Lion and has native UAC2. there is no straight up financial incentive (though of course someone could argue just having it is, but as i said, nothing to hide there), for me just having a cool toy to play with that was unheard of at the time and happened to suit the direction i was already going quite well, hard to turn down. especialy as like you i have tried a number of USB interfaces and found them a bit wanting, always missing some aspect. but its only one of the options i'm testing and i'm still praying for a native mac version of allocator

IMHO, this is a very interesting project and it'd be nice to be improved further (by comparing/measuring clocks or something) than getting stuck on which transformer to use.

agreed!! I do believe thats what he got the little pcbs for and why the clocks are socketed, i dont think its to improve performance =P so yeah on with it Ian!! :whip:
 
Last edited:
Unless you have checked the system and particularly the scope I would not put too much analysis into the fast rise time abberations. It possible that the overshoot and preshoot are even in the scope itself (another ADC with many of the familiar digital issues). One is faster than the other but that may be as far as you can tell.
I have had to invest in a lot of stuff to be sure my instruments aren't lying to me.

Agree!

Ian
 
Here is a 3.3V regulator I'm using. I'm still verifying its performance. The sim shows .7 nV/ rtHz which is really really low.

A variation of the source:
http://www.diyaudio.com/forums/digi...timate-weapon-fight-jitter-3.html#post2710526

With European transistors BC550C and BC560C.
Maximum 100 mA 5-9 V input.
Adjustable R2 (Vout 3.3 V. depends to each Jfet).
An externally hosted image should be here but it was not working when we last tested it.

Noise 0.2 nV/ rtHz vs 0.7 nV/ rtHz.
Impedance 55 mOhm vs 250 mOhm.
LED voltage reference avoids potential problems of instability when the TL431 current is below 20 mA.

More mA: 5 parallel BC560C and R1=1 Ohm. 500 mA. 5-9 V input
0.2 nV/rtHz noise, 32 mOhm impedance.
 
Member
Joined 2004
Paid Member
I have a few questions on your variation.
1) How did you determine that the BC550 is 12 dB lower noise? I have yet to see any single device much lower than .3nV/rtHz and that was an unobtainium JFET. I have measured a number of bipolar devices and have not seen such low noise. I'll try to get some and see what I measure.
2) Depending on the load device's sensitivities the increased variations in voltage vs load and voltage vs temp may be fine in using the circuit open loop. For precision crystal oscillators even small variations in the supply can be measured.
3) Where do I find out more about "instability" in TL431's below 20 mA? I have never seen that and the spec sheet has them working fine to .5 mA.
 
This is an LTSpice simulation data with Philips BC550C Spice.
I had simulate your circuit also with theoretical results of 0.7 nV / rtHz.

I thought the key was the topology, not the transistor. I have not chosen BC550C precisely because of its low noise. My intention was to test the transistors and LEDs that I have at home. I have no LT431 or 2N3904.

After your question I tried now with BC549/550 spice model I have just received by mail from FAIRCHILD. And result 1.5 nV / rtHz. 8<((

I dont know if Philips BC550C is better or if there is another more suitable transistor on noise. In datasheet 2N3904 and BC550C have similar characteristics on noise.

About variations with respect to the reference voltage load, takes not reference at output, and output impedance is less, so it should affect less to load variation. But input noise affects more.

LT431 instability has been discussed in another thread diyaudio, is specified in datasheet in http://www.ti.com/lit/gpn/tl431 Figures 15 and 16. And take out reference to then act on the output capacitor through.

I thought that was the reason for putting the 0.1 uF capacitor C9 of your circuit. LTSpice simulations did not see any effect of the condenser.
What is the reason for putting that capacitor C9 in its source?

Best regards.
 
Last edited:
Member
Joined 2004
Paid Member
The topology of the pass circuit is the key. The TL431 is only to provide a stable DC voltage. The cap from sense to the positive terminal was to stabilize the device (sort of brute force) but in this application it takes seconds to stabilize the voltage which is fine.
It actually has no cap loading (A-K) at all and is between two moderately high value resistors. It should be very stable.

The number you are getting from the sim now makes more sense. The .7 nV numver is almost a practical limit. keeping the npn from passing current except the bias current keeps its noise low. The common emitter circuit of the pnp is the lowest contributed noise option for that stage (an emmitter follower would have higher noise).

If a stabilized voltage is not critical a resistor divider for the base reference works well. However the output voltage will vary with load, line temperature and from transistor to transistor. the TL431 is the quickest and cheapest solution I know (and I have those on hand).

Measuring this is very difficult since its noise is lower than many MC phono stages. I have a transformer coupled thing I built but it has too much hum pickup and I don't have time to sort out the problem right now.

Enjoy.
 
With my TL432 spice model, comes to 30 Hz instability that produces a noise peak of about 120 nV / rtHz when put 0.1 uF C9. Removing raises instability but the low frequency noise below 10 Hz to 4 nV / rtHz. Is from there I had comment on this instability. What model uses? Can we share models?

In simulations can be similar used transistors type BCXXXX and 2NXXXX.

Do you think the BC550C be good for the two sources?
Is the 2N3904 is better and I have to make the effort to get it?
Is there another that can be even better than these two?

I have measured several sources and regulators. The lowest noise and input noise rejection, especially at low frequencies has been Salas source.

The idea is to test this now as a miniature to Trident regulators to lower the power output of Salas 5V to 3.3 V. for the clock with very low noise.

I am going to mount the two topologies as a source for Audio Clock Widget. I Will attempt to measure noise source and J-test jitter at the audio output.

I was wrong and wanted to have written all this in Audio-Widget tread..

Best regards.
 
Member
Joined 2004
Paid Member
I am traveling right now so I can't verify anything but I can say that I saw no evidence of instability in the regulator. The 2n3904 is not too special and there are many equivalents.
The Pease dictum still holds- don't trust sims.
You won't be able to see the noise at this level without something like a paralleled fet preamp. Jim Williams published one for looking at regulators 20 dB noiser than this. I built the transformer based fixture since it is the only way to get that low. Now to get the hum pickup out of it. . .
I doubt you would see it in something like a JTest since other stuff will dominate the measurements.
You would need a phase noise tester with a really low noise floor to see its effects.

Sent from my T-Mobile G2 using Tapatalk
 
I have limited means to make absolute measurements. But try to make a comparison between an Audio Widget unmodified and other modified with different power sources. And comparative measures wiht Hiface , Buffalo DAC and others ...

Do you Know any Internet address information from this Jim Williams circuit?
 
Member
Joined 2004
Paid Member
Here is a link to his circuit: http://cds.linear.com/docs/Application Note/an124f.pdf
Its a real bear to build and its not really low enough to get out of the way. I used a "geoformer" a type of transformer used for interfacing to seismic sensors. The one I have has a 35:1 turns ratio and a low DCR on the source.

However with such a high sensitivity it takes little hum field to swamp everything else out. I think its an interesting problem to solve but it won't be a 5 minute effort. And it needs to be cap coupled. The $100 wet slug tantalums are not in my budget. Verifing the gain and noise is an exercise as well since the input Z is pretty low.
 
Metal base integrating everything together

I designed a very simple metal base by Solidworks trying to integrate everything together. A friend of mine machined it by a CNC for me. Although this one is not as good as a real case, but it’s already much better than the wood base.

With the FIFO board and the double XO clock board hooked up on the metal base, it become a I2S FIFO with possible low jitter secondary clocks and automatic Fs switching function; And by adding another S/PDIF interface board, it become a working S/PDIF FIFO.

The S/PDIF interface board on the picture is a new upgraded version with balanced AES/EBU XLR output, because I found connecting the output of the S/PDIF FIFO into a DAT or other professional digital audio devices is very interesting. But personally I like the coaxial cable a bit more.

Connections of the I2S input and output bus between the FIFO board and the S/PDIF board could be either 7 pins PH2.0mm cables or U.FL cables. I like using U.FL cables connecting I2S buses because they come with better signal quality (also bit more expensive), although some DAC is only sensitive to the jitter from the master clock.

Now, I’m trying to integrate my other digital audio sources with the S/PDIF FIFO into my system, such as Squeezebox Duet, Apple TV II, Foobar2000 and Mega CD storage. Those are very pleased experiences. For example, by buffering the digital output of the Squeezebox Duet with a S/PDIF FIFO which equipped with CCHD-957 low jitter clocks, the Squeezebox is no longer the one it uses to be. I think this is one of the best ways to boost the audio quality without opening and modding those devices.

Ian
 

Attachments

  • MetalBase.JPG
    MetalBase.JPG
    262.1 KB · Views: 1,365
  • I2SFIFO.JPG
    I2SFIFO.JPG
    475.8 KB · Views: 1,366
  • SPDIF_FIFO_WithPH2cable.JPG
    SPDIF_FIFO_WithPH2cable.JPG
    540.5 KB · Views: 1,367
  • SPDIF_FIFO_with_UFLcable.JPG
    SPDIF_FIFO_with_UFLcable.JPG
    515 KB · Views: 1,321
  • SPDIF_FIFO_Squeezebox_AppleTV.JPG
    SPDIF_FIFO_Squeezebox_AppleTV.JPG
    357.4 KB · Views: 1,282
Last edited: