Hunting for noise

[Nick]

My plans to hunt down the noise problem in my system got off to a false start with the purchase of an Oscilloscope that turned out to be no better than a toy.

( more here  http://www.phasure.com/index.php?topic=2404.msg27276#msg27276)

Six months later I'm getting started. A new oscilloscope, a Rigol DS2072, arrived a couple of days ago. I have to say the Rigol is an absolutely excellent bit of kit. I also bought a PC attached USB logic analyser a few weeks ago with 8 digital inputs and 2 analogue traces. The cool thing is that the logic analyser decodes the I2S serial protocol which allows me to view the bit clock, word clock and data values being sent into the DAC section of my NOS1, The bits and bytes look good too after Peter has applied XX magic to them  So tools at the ready !

I spent a few hours today measuring round my NOS1, PC and gainclones with the Rigol scope to get a feel for what may be going off that could cause poor sound quality. Some VERY surprising stuff is coming to light. I want to spend more time working up solutions but a couple of things that are interesting so far.

1) there is a LOT of noise in there.
2) I can clearly see activity from the PC (disk drive and PC psu) causing noise on the NOS1 outputs (so the PC does have potential to electrically influence analogue outputs ??)
3) after applying a few ferrites to the system aimed at reducing HF noise my sound quality has taken a leap forwards. The ferrites are treating the symptoms only, dealing with the source of the noise I'm seeing may be a lot harder.
4) I use a "none standard" 22mhz oscillator (clock) in my NOS1. The sound is very good with the clock but the wave form of the clock looks like it could be better. I think I'll need to investigate how to achieve better line matching with the clock input of the USB interface.

It's nice to be making some headway and to be using a scope that can be trusted !

Regards,

Nick.

[PeterSt]

Now Nick, don't tell me that you didn't try to play music with the DAC section off as a first ?!

(this will be vague to all, sorry)

Peter

[Jud]

Nick, great to see someone being serious enough to buy equipment I'd love to have, but for now isn't in the budget.  Will be happily following along, reading about your "discoveries."

[Nick]

Now Nick, don't tell me that you didn't try to play music with the DAC section off as a first ?!

(this will be vague to all, sorry)

Peter

Peter hi,

I have not tried that yet although the balanced gainclones are a better in that respect so I am not sure they would perform 

I think I may know what is allowing the break through at -144db volume 16x up sample and the poor SQ to happen. How many times have I said that, haha. The plan is to try to repeat the measurements and check the effect of the ferrites again. If it checks out as before I will drop you a pm, I may need some help 

Regards Nick

[Nick]

First finding - How to encode an audio signal in an I2S data stream.

A quick recap

I have been having a problem with my audio system for a year and a half. There are two main symptoms:

1)   General loss of resolution increase in harshness collapsing of sound stage and dynamics when playing at high (16x) up-sampling rates (linked to when 2) is present) and,

2)   When Playing music at a -144db XX volume setting, I can hear a distorted version of the music playing at low levels through my speakers. What is surprising is that the DAC converters and IV section of my DAC can be turned off and yet still this breakthrough music can be heard.

Trying to approach this my thinking was that symptom 1) is quite general and none specific, it could be almost anything causing it so its difficult to test even with my new test equipment on hand.
Point 2 however is quite interesting. Key here is the diagnostic assumption that if I could hear the music breakthrough even with the DAC turned off then something in the digital domain upstream was decode data to an audio frequency music signal. But how was this happening ?

Main suspects were:

1)   Some unintended DAC process in the PC (there are DACs in the PC but there shouldn’t be working …). I decided this was un-likely following a few tests and discussions with Peter.

2)   I have worked on USB to SPDIF interfaces with build in DACs in the past. Might it be that the NOS USB interface has a DAC on board and this is unintended decoding the music data. Peter kindly confirmed that there is no such feature on the NOS1 USB interface, so this was ruled out.

3)   Is the USB stream from PC to DAC somehow modulated by the music signal ? Could possibly be the case but not sure how. If necessary I could can test with my logic analyser

4)   Is the I2S data stream that links the NOS1 USB interface card to the DAC might somehow be modulated by the music signal ? This stream of data is very close electrically to the output of the DAC, there are many reports of I2S being noisy on the net, and there is the possibility of timing problems with I2S signalling so this is where I decided  to start looking.

Peter has been very helpful and we have been sharing ideas and experiences relating to I2S observations over the last weeks. Amongst points Peter raised was that the 1704 DAC chip uses 2’s complement binary data encoding to pass music data over the I2S data stream into the DAC.

When looked into the key here is that bit values and therefore the electrical “on” and “off” values of the I2S data near the zero vale of the musical wave form look very different when they are passed down the I2S data connection.

For example:

0 (dec) = 0000 0000 0000 0000 0000 0000 (24bit Binary) so electrically for this music value the I2S data line is “off” at 0 volts to send a 0 value. Likewise positive low values are mostly zeros (offs).

-1 (dec) = 1111 1111 1111 1111 1111 1111 (24bit Binary) so electrically the I2S line is “on” at 3.2 volts to send a -1 value.  Other negative music data values just below 0 are also mostly 1s (ons).

A couple more useful bits of information that turned out to be relevant:

• I2S data is sent at the logic gate values of the sending and receiving chips so in the case of my interface that’s 0v = an off bit and +3.2volts = an on bit.
• I estimated the peak to peak voltage amplitude that would be needed at my DAC’s output XLRs to generate the breakthrough music sound level I hear played whilst at the XX volume setting of -144db to be in the rage 100uv to 500uv. (from an estimate that the volume heard is about 50db at 1m, played via 103db/w/m loudspeakers and a 36db amp).
  

So the I2S carrier of 0-3 volts square wave signals at a carrier frequency (data bit rate) of up to 22mhz per I2S line  and there are two I2S data lines in use within the NOS1. Is seemed possible that the breakthough audio signal of 100-500uv could be coupled to the output of the DAC even with the DAC chips turned off. Now time to look for how an audio frequency signal could be encoded into the electrical conditions of the I2S data lines.

I obtained an 8 channel logic analyser (which decodes I2S data) and a 200Mhz oscilloscope to probe the I2S data lines with music playing at -144db. I was amazed to see the following emerge.

Traces taken with logic analyser.

Key of the trace lines in each picture.

• Top trace line is one of the two I2S lines in the NOS1 with the software set to decode the music word values being passed to decimal values. Note that that since this is tws compliment data that the music value of -1 is decoded to Dec 1677215 (bin 11111111 11111111 11111111).
• The next trace line down is the raw I2S data on the NOS interface so low is 0 volts and high on this trace a 3.2v condition.
• The next line down is the word select clock which changes state every 32 bits word sent on the I2S data lines.
• And the final  line down is the bit clock signal.

First trace
1 khz test tone recorded at 0db played with XX volume at -144db not upsampled (44.1khz).

This is a close up of what happens when the music wave form values (in this case a sine wave attenuated by -144db) moves from a positive in the wave cycle to a negative part of the wave cycle. Note that when the music wave was above the zero point the wave is attenuated to the binary value of 0, so the electrical state of the I2S line is mostly 0v. However when the sine wave was in the negative cycle the attenuated value never gets to 0 it stays as -1 and due to the two compliment treatment of the I2S data the electrical value of this -1 data is a state of mostly +3.2volts.

I was looking in my system for a misbehaving semiconductor acting like a diode to demodulate the data transmission, but the actual I2S line and Twos Complement state values themselves are creating an electrical state on the I2S data line that corresponds to zero crossing points in the music wave form ! oh boy 

So now lets look at zoomed views of the trace so that you can see a longer duration of the sine wave with more transitions of the music sine tone across the zero value point.

Humm, the electrical condition of the I2S data line is now a good approximation to a square wave at the 1khz frequency of the music sine wave. The square wave has a low value of 0v when the original audio sine wave signal was of a value below zero and a high value of 3.2 volts when the audio value was below zero.

Bloody hell QED  ! the audio signal IS being demodulated into the I2S data transmission voltages.

From here it is a short step. The modulated data (I have looked at many many more measurements including normal music and its facinating watching the demodulation take place, the audio looks like Pulse Code Modulation centred on the zero crossing point as the volume us raised from -144db upwards) is being carried at up to the bit clock frequency on a square wave of approaching 22mhz. This frequency of carrier and its overtones can couple with near by traces / connections power rails etc. I need just 100uv to couple to the DAC outputs and that plays the breakthrough sound – which is what is happening.

Continued in next post

[Nick]

Continued 2 of 2


So a few throughs on this process.   

All DACs that use I2S inputs and twos compliment encoding for input data will encode audio in their I2S data line electrical transmission (some DACs use straight binary encoding and will not suffer). This is not a special condition of the NOS1 although the high data rates used by the NOS seem to complicate matters here in my system. The problem is at its most noticeable when the I2S transmission rate is high (eg when up sampling at 16x) and both 12S lines are active together. Playing at -144db highlights the modulation to the maximum degree but it is always happening as music plays. What I can say is that the amount of breakthrough at -144db that I hear is strongly linked to how good that quality of normal music replay will be in my system. So far I do not understand the coupling of the I2S line to the outputs of the DAC but this is what I have to work on next.

Finally, I must thank Peter who has spend a lot of time exchanging mails with ideas and information about the NOS and the encoding of the I2S lines. We had a very interesting conversation earlier this week using webex to share a number of analyser traces and the conversations are really helping me to plan next steps which will be about controlling PC and DAC noise levels.

Nick.

[PeterSt]

Hey Nick,

Yes, what a travel this has been. But allow me to make a tiny correction because I am too proud on myself ...

Peter has been very helpful and we have been sharing ideas and experiences relating to I2S observations over the last weeks. Amongst points Peter raised was that the 1704 DAC chip uses 2’s complement binary data encoding to pass music data over the I2S data stream into the DAC.

So ... I am really too proud on myself to skip the small part that at that stage I finally could tell what was going on, and how without a DAC music can be heard from i2s itself.
So, with your analyser under way, and for me too after 1.5 years or so throughput time and countless emails about this totally not understandable phenomenon, I could tell that the zero crossing you so beautifully explained for next readers, happens like this :

Zero crossing (voltage goes from minus to plus and back) happens at the lowest frequency only;
This implies a voltage jump - the highest possible - because it's going from all zeroes to all ones in the DAC.
It was my assumption that even with the DAC off, this could leak through because of the relatively high current needed to do this.

At this stage I needed to measure the output of the DAC with the DAC section off, while your analyser was under way.

Do notice the importance of Nick's telling that the lower the digital attanuation, the more the music was audible.
Nice puzzle.

Now, with the help of the logic analyser, we can see that this current needed (the voltage jumps) already happen in front of the DAC, with at this moment the jury still being out to the leakage through the outputs. But :

What I by then (this week) could reason out is that the lower the level of the music (like -144dB is ultra low) the lower the level of the wave of that low frequency, and the LONGER the time period the higher frequencies riding on that lowest frequency (which is how it works) are able to cross zero themselves. This is how the more attenuation, the better the representation of the music becomes, because the more higher frequencies imply the zero crossing.
Compare this with 0 attenuation, the frequency (time distance of the lowest frequency wave) not changing, while the zero crossing is close to vertical and there's no time for higher frequencies to go back to plus once the low frequency crossed zero to minus.

Next, but actually right before the "analysis" from the last paragraph, it was totally 100% Nick who could explain by means of the graphs shown in his previous posts, that this is a pure i2s thing, while I anticipated an in-DAC chip thing. So, the "pulse width modulation" I named it by looking at the i2s behaviour (I think Nick named it somewhat differently) has a life of its own, and Nick pointed that out so SUPERBLY.

At writing this I feel like a couple of guys won a Nobel prize. Really.
It wouldn't be the invention of the century perhaps, but I am pretty sure nobody ever saw this.
What a victory.

For me this definitely means how further improvements on a DAC are possible and right away it is in some design already.

For Nick it is the path to eliminate this noise leaking through, which really should not happen. It doesn't happen to me, at least not audibly even the slightest, but what can happen with Nick can happen to anyone, thinking about external factors.

Great stuff !
Peter


PS: I edited out one aspect from your post which was irrelevant (and therewidth wrong), which correction I hope you will grant me.

[PeterSt]

And Nick, might you still have that picture, IIRC the -108dBFS attenuation shows the Pulse Width Modulation at the bit level (or otherwise it was the -124dBFS file). I think what you showed in your post shows something at the audio word level, which does not justify the PWM idea.

One other thing to add from my side :
The zero crossing at -144dBFS (which is what Nick showed) should not be there at all, but is. I dedicate this to something wrong in the USB driver, knowing that XXHighEnd really throws out zeroes only at that level of attenuation. I longer ago saw this in my own analyser too.
This is not important for the whole story but should be solved. Just saying ...

Peter

[Nick]

Peter hi,

Sorry it absolutely not my intension to understate the amount of help and support you have provided. All of the guidance you have given to help me with the noise problems (proberbly 100s of emails :-) ) and the conversations we have had leading up to capturing this view of the I2S data. I decided to look at the I2S bus but without this guidance and your generosity with information and insights these views of I2S would not exist. You know that there is much that you have said to me that is not for discussion  that would immediately demonstrate the depth of your understanding of what happens across the end to end DAC system 

I also agree it is unlikely that others have not seen this demodulation of audio data in action, and I also feel like we reached a goal together in understanding this small element of a DAC operation. I think this will prove useful.

And Nick, might you still have that picture, IIRC the -108dBFS attenuation shows the Pulse Width Modulation at the bit level (or otherwise it was the -124dBFS file). I think what you showed in your post shows something at the audio word level, which does not justify the PWM idea.

One other thing to add from my side :
The zero crossing at -144dBFS (which is what Nick showed) should not be there at all, but is. I dedicate this to something wrong in the USB driver, knowing that XXHighEnd really throws out zeroes only at that level of attenuation. I longer ago saw this in my own analyser too.
This is not important for the whole story but should be solved. Just saying ...

Peter

Interesting about the USB driver convering zeros to ones. The logic analyser can decode USB data so I could try to take a look at what is passed on the d+ d- lines.

Here are the traces that show the PWM effect of the positive binary count and negative binary count (in twos compliment) on each side of the test tone zero point. The zoomed really show the effect nicely.

For folks  looking at the traces, concentrate on the PWM appearance of the brown coloured trace labelled “IS2 2”.


Trace details
Sine Test Tone 1khz 0db
Played in XX at -108db vol
DAC 24bit 88.2khz

Best Nick.

[PeterSt]

Sorry it absolutely not my intension to understate the amount of help and support you have provided.

Hahaha, as I could expect, nothing could prevent you from now thinking something like that. But no, this was not my "message" at all. I didn't have any message. But what I wanted to make clear is that it is of such vast importance for me that I could finally reason out that such a phenomenon - and without measuring - could make music. This after all the so many emails with reasoning I could not. And so I dedicated 50% of the "Victory" to myself, because I always like to apply (new) theories and after that let them work (no matter what).

Like your first post about it, for me too is was a "Finally ?!". Could be hard to understand for outsiders, but after 1.5 years of (regularly) thinkering about how the heck music could emerge without D/A converter it really is something which was achieved. For history, here's a part of our 100s of emails throughout the process; this is what I wrote last July 10, and btw I can see from the text that you had the logic analyser by then :

Cautious conclusion : there really can be something in your "breakthrough" from the one to the other i2s channel. But it needs real data, and don't ask me how you can judge that through your digital probes. Hopefully you can (and hopefully you have that real i2s package of which I said you wouldn't need that ? hahaha).

Still my analyser keeps on showing the test signal (of each frequency) no matter how much I attenuate (FYI : stays as high from off -108dB all the way to -144dB).
I can not see that on the scope part, but the FFT digs it out.
I never tried measuring this with my DAC section off, but by now I dare to bet a few things ...


When some interestee, later, passes along this topic, I expect those with the knowledge to say "well, so what, this is just i2s data we see here". And this is correct of course, and nothing different from my sort of reluctancy to not needing an i2s decoding package as we see (implied) quoted in the first paragraph above. So, while my theory was hung up to glitching in-DAC (which would be a sheer R2R process in our NOS1 case), it was still you coming up with the data after which we could see that this glicthing is actually an inherent i2s process. And THAT is new.

So to work this out somewhat more from the technical side of things - if not already done by yourself in the first post - ... when an R2R chip crosses zero and say a +0V for zero turns into a -2V for -0.0001 (just one smallest value lower than 0) there's a large current implied and this is named a "glitch". This expresses as an AC voltage (read : music) spike of a couple of 10 uV, depending on the D/A chip (btw the PCM1704 says it isn't there, which would be true because I can't see it because of the relative high level of noise).
With i2s though, this is 3.3V going to 0V and the other way around, thus even more than in-D/A chip.

So again from the quote above - now second paragraph - I already had worked out that something has to be going on, because I had seen that beyond the resolution the D/A chip can resolve (because all drops into the noise) I could see a very persistent normally looking test signal of any frequency remain, no matter how much the attenuation. This by itself is a story in parallel and about seeking a bug in some design which bug could not be found and I gave up and layed it aside.
Now, today, it is totally clear to me that whatever the frequency of the test signal, and no matter how low level it is, there is this glitch in i2s which happens at zero crossing and which of course is 100% equal to the frequency itself.

Because this frequency keeps on sticking above the field at a standard level (something like -120dB IIRC) it is just the glitching energy of i2s I see (at a noise level of -143dB). So, this is about the few tens of uV I talked about, and it will never go away. With real music the same will happen, and each zero crossing, which can be for the higher frequencies which modulate on the lower) this will be just the same, and if only the level of the music is low enough, this other source of "music" will become profound, but is distorted. It is distorted because only when zero is crossed it can briefly do its work for the higher frequencies, while for the lowest it will be rather pure with the notice the level of it can't change (because it's always implied by the 0V-3.3V current needed).

So, learned another thing ...
What's next ?

Peter
 

[pedal]

Wow, hats of to you guys.
I will read the full thread later, but it seems you have made a world discovery!

The i2S is the internal interface/signal format within all integrated CD players, right?

Does this mean that ALL CD-players suffer from this kind of distortion?

[Nick]

Wow, hats of to you guys.
I will read the full thread later, but it seems you have made a world discovery!

The i2S is the internal interface/signal format within all integrated CD players, right?

Does this mean that ALL CD-players suffer from this kind of distortion?

Pedal hi,

Yes I2S is a the line standard for sending digitally encoded music. In out case it sends the data to the DAC board of the NOS. The is a data line with in our case twos compliment encoded data being sent. A word select line which triggers 32bit music data chunkes to be sent to left and then right channels and last of all a bit clock which "times" each bit of the 32bit data words.

The NOS is rather more clever than a "normal" DAC in how the data is handeled but essentially I2S is what moves the data towards the DAC chips.

All CD players the use I2S to move data coded in twos compliment binary form will experience the decoding of the audo signal into the electrical conditions of the I2S data line. Depending on a number of factors this may or may not effect the DAC performance (a low res slow speed dac with poor noise floor will probably not suffer  but that is not the NOS).

Some DACs use straight binary encoding rather than twos compliment and will not suffer BUT even these may encode music onto the I2S line, but rather than this happening when the music crosses the zero point as with I2S the straight binary the encoding will happen as music waves get towards minimum (more zero bits sent)  and maximum (more 1 bits sent) music values. This is likely to be much better however as the signal level is in the region of 0db so the encoded noise should be drowned anyway.

When Peter is sweating the last grams of performance for us this is very important stuff.

Cheers Nick.

[Nick]

Peter hi,

>>What's next ?

With the aim of eliminating noise here I was thinking.

Probe the I2S into the 1704s and check data against where I am currently sampling. Also check analogue I2S conditions into the 1704s this gives the full picture dome the the converters just to be complete.

Check the D+ D- analogue conditions of the USB link to see how they look and maybe decode the data but not sure of I'd work out the frame structure (proberaly not that important to do).

Build a field probe to go looking for devices / lines that are emitting the I2S encoded data and where coupling to unwanted areas may be happening.

See the last post to Pedal above on binary encoding I think it may also encode the elecrical I2S line in phase with the music wave form but much better behaviour at wave values in the region of zero than with twos compliment.

-----------

On another subject, Paul and his wife were with us for a very enjoyable day today. Whilst tinkering with my music system we put Paul's PC on the front end of my NOS and SQ was so absolutely brilliant !  I could stop hear with the noise quest it was so good, it was mouth open time for both of us :-). The -144db noise is there as expected and music still sounded better at 8x up sampling still (hence the intent to look at data reaching the 1704s just in case).

I suspect I have a duff PC somehow ! So something else concrete to look into. Right now Paul and I were thinking I should build a PaulClone PC, there is something magical in the sound of Paul's machine both at his house and at mine.

Cheers,

Nick

[Scroobius]

Just a note to say that I do not have any noise breakthrough problem here at my house. Well by that I mean that if I play music at -144db there is no breakthrough at all that I can hear in my system at my home. It does not mean that the IIC problem is not happening here though - I would need Nick's scope to check further and that is something we may do when Nick's comes here in a few weeks time.

However yesterday we tried my PC at Nick's house and the breakthrough problem happens with my PC. It also happens with my NOS1 and (not surprising) with my old single ended GC amp which we have tried on a previous visit.

So what is happening at Nicks house that is not happening at mine?

Very weird indeed.!!!

Paul

[boleary]

Hey Nick didn't you post some time ago that re-enabling devices in your BIOS had a positive impact on SQ? Here that is definitely NOT the case. Disabling all Onboard Devices in the Advanced/Onboard Devices Configuration tab is essential to the incredible sounds I've been hearing.