I have spent quite a lot of time experimenting with several modes of i2s connection, grounding and wiring and here is the final conclusion that I came to:
I have experimented with the following 3 methods of connecting converter to DAC via i2s:
1. TTL with only one ground wire and 4 signal wires (data out, lrck out, bck out, clock in)
2. TTL with 4 ground wires running in parallel with 4 signal wires
3. RS-485 using flat ribbon cable
And soundwise the winner is #2. TTL with only 1 ground wire has sound that has a small amount of noise which makes it very hard to listen to for any continuous period of time.
RS-485 sounds better than TTL with one gound wire but it has a small amount of noise as well that is quite difficult to listen to for a long period of time.
Also I want to say that I conducted the following experiment with RS-485. I cut the flat ribbon cable into 4 balanced pairs whereby 1 balanced pair was used for each i2s signal (data out, lrck out, bck out, clock in) and then I twisted wires in every balanced pair. So in efect I got 4 twisted pairs for RS-485.
I was unpleasantly surprised because it turned out that 4 twisted pairs sound worst for RS-485 than just standard ribbon cable. Why is that so, I don't understand.
Another experiment that I conducted was wrapping each balanced pair in a copper tape and connecting all these copper tapes to the DAC's ground. This configuration produced very bad sound with some kind of distortions which are very difficult to listen to. Why is that so? I can't find any explanation in view of the fact that as far as I can see people like to use standard HDMI cables or custom made HDMI cables for balanced i2s and such cables are sold at very steep prices.
Could anyone explain this phenomenon?
I have experimented with the following 3 methods of connecting converter to DAC via i2s:
1. TTL with only one ground wire and 4 signal wires (data out, lrck out, bck out, clock in)
2. TTL with 4 ground wires running in parallel with 4 signal wires
3. RS-485 using flat ribbon cable
And soundwise the winner is #2. TTL with only 1 ground wire has sound that has a small amount of noise which makes it very hard to listen to for any continuous period of time.
RS-485 sounds better than TTL with one gound wire but it has a small amount of noise as well that is quite difficult to listen to for a long period of time.
Also I want to say that I conducted the following experiment with RS-485. I cut the flat ribbon cable into 4 balanced pairs whereby 1 balanced pair was used for each i2s signal (data out, lrck out, bck out, clock in) and then I twisted wires in every balanced pair. So in efect I got 4 twisted pairs for RS-485.
I was unpleasantly surprised because it turned out that 4 twisted pairs sound worst for RS-485 than just standard ribbon cable. Why is that so, I don't understand.
Another experiment that I conducted was wrapping each balanced pair in a copper tape and connecting all these copper tapes to the DAC's ground. This configuration produced very bad sound with some kind of distortions which are very difficult to listen to. Why is that so? I can't find any explanation in view of the fact that as far as I can see people like to use standard HDMI cables or custom made HDMI cables for balanced i2s and such cables are sold at very steep prices.
Could anyone explain this phenomenon?
Without a 'scope on the digital signal lines, its guess work, but its likely digital signal quality or clock skew. By "clock in" do you mean MCLK? Normally the clocks all go the same direction so something seems odd with your description. By TTL do you mean 5V logic? TTL isn't used these days, logic is typically 5V, 3.3V or 1.8V CMOS
Note that high speed logic signals aren't designed to travel distance. RS485 are, but if your RS485 converters are slow and have a lot of latency and jitter its not a good choice. LVDS would be one method of sending high speed logic signals over moderate distances, and less power hungry.
Note that high speed logic signals aren't designed to travel distance. RS485 are, but if your RS485 converters are slow and have a lot of latency and jitter its not a good choice. LVDS would be one method of sending high speed logic signals over moderate distances, and less power hungry.
You know RS485 is polarity-sensitive? Did you terminate each pair?
I have forgotten to mention that my DAC has synchronous reclocking, so I believe latency and jitter introduced by RS-485 should not be an issue at all. My DAC has ADM1485 as RS-485 converters.
Yes, of course I know that and each pair is terminated at both ends (at the DAC and converter)
With I2S its not just a matter of individual signal integrity, there is also the issue of common-mode noise emanating from the source. If the layout results in common-mode noise flowing in the analog groundplane then there's the chance of common ground impedance coupling.
I am not the one who designed my DAC and converter but I believe that the guys who designed my DAC and converter are quite knowledgeable and professional.
The explanation is very simple. It's just that the converter and PC is synchrosied by the masterclock in the DAC. That's why masterclock goes in the opposite direction to the signals coming out of the converter and into the DAC..
Given i2s is digital, it's possible to simply send digital test data, then sample the output and then compare the input and output data to get a objective test.
If the test data out is the same that is going in.. then there's another issue at work. Possibly noise. If you have different data then there's a problem with the data transfer itself.
I got a 27dB drop just by adding a digital isolator chip between my ADC and the STM32. That dropped further by clocking separately with a 1:3 fan out line driver. The key here was the drop in noise. Switching to a -100dB close in CCHD oscillator made a minor difference to 'jitter' compared to actually sorting out the noise from the power supplies and isolating.
If the test data out is the same that is going in.. then there's another issue at work. Possibly noise. If you have different data then there's a problem with the data transfer itself.
I got a 27dB drop just by adding a digital isolator chip between my ADC and the STM32. That dropped further by clocking separately with a 1:3 fan out line driver. The key here was the drop in noise. Switching to a -100dB close in CCHD oscillator made a minor difference to 'jitter' compared to actually sorting out the noise from the power supplies and isolating.
My DAC is divided into 2 sections: digital and analogue and they are isolated by ADUM IC. Converter is connected to the DAC in the digital domain. So I believe there's no point in talking about the analogue groundplane.
If you have no desire to investigate analog groundplane contamination, that's entirely your prerogative.
I used SMA connectors and 12cm long miniature screened cable to connect my DAC to a modified CD player.
I couldn't detect any noise in the audio.
I couldn't detect any noise in the audio.
In my case the sound source is a sound card in a PC with AES/EBU output going to the SPDIF - I2S converter and then to DAC. Could it be that some noise was coming out of the PC with the signal? On the other hand when I experimented with optical connection between the sound card and converter I did not hear any advantage of optical connection over coaxial connection.
I think I would probably be less quick to reject abra's hypothesis.. I found ground noise to be one of the main issues as it causes the detection of data (ie noise inducing 'jitter') to be corrupt. The oscillator timing and the data detection timing both are susceptible to that noise.
I had a 24.xMbit/sec i2s feed but used 250Mbit/sec TI digital isolator. The ADUM410 (from just searching) has a max bit rate of 10Mbit/sec. So I assume that's a slow i2s bit rate being sent through (unless there's another ADUM model that has a higher bit rate). Using a low bit rate digital isolator simply means a higher quantisation error. In my case I use a 10:1 to reduce the quantisation error.
I had a 24.xMbit/sec i2s feed but used 250Mbit/sec TI digital isolator. The ADUM410 (from just searching) has a max bit rate of 10Mbit/sec. So I assume that's a slow i2s bit rate being sent through (unless there's another ADUM model that has a higher bit rate). Using a low bit rate digital isolator simply means a higher quantisation error. In my case I use a 10:1 to reduce the quantisation error.
Well it's just your assumptions. The question is why an analogue contamination (if this is the case) could occur when all I did was just use 3 different methods of connecting converter to DAC? And the analogue section is isolated from the digital one by ADUM.
Again the same question: why shoud it be relevant when all I did was just use 3 different methods of connecting converter to DAC? And the analogue section is isolated from the digital one by ADUM.