Building the ultimate NOS DAC using TDA1541A

Interresting , what do you think will be the issues having a to quick slew rate ?
Sorry @fabrice63 for late reply. I missed this one.

I was informed that:
"faster than necessary" flanks
and
"higher amplitude than necessary"
signals

will cause noise contamination in substrate which will leak globally to the entire chip (shows up at the output + affect other functional blocks).

"Stopped clock" operation eliminates BCK related noise (great idea) at the moment of conversion (when it matters most), but does not eliminate other contaminants.

I implemented stopped clock operation + Slew Rate attenuators, but sonically it was equivalent to not using stopped clock at all (no sonic benefit) - something that others appear to have also noticed.

I would say it is a really great idea, but attenuators lead to better objective and subjective results.

Cheers
 
  • Like
Reactions: fabrice63
I was informed that:
"faster than necessary" flanks
and
"higher amplitude than necessary"
signals

will cause noise contamination in substrate which will leak globally to the entire chip (shows up at the output + affect other functional blocks).

Yes. The input transistors are PNP in a specific technology, that creates a lot of capacitance to the substrate.

The PNP transistors load with a resistor from emitter to +5V and most likely connect the collector for to -5V.

There is another NPN transistor that follows, collector to +5V, level shifter diode chain in the emitter also with emitter resistor to -5V. After that is a standard CML/ECL input for CML/ECL(ogic) between DGND and -15V.

Very standard for late 1980's high speed circuitry.

The circuit creates several problems.

Currents flow in +5V & -5V from these input circuits. The greater the input swing, the greater these current. This is likely the reason why HtP insisted on a direct RF decoupling path between +5V and -5V.

Internal clamping limits the swing at input PNP transistor emitter to 0.6V...2.4V for 0V to 1.8V (temperature dependent) input. At higher input (than ~ 1.8V), the Input PNP transistor reverse biased, meaning to will be slow to react to a falling edge.

The best range for the input, if my "TTL 2 ECL" hypothesis is right, is around 0.9V to 1.8V, which matches well with practical observation. Using 1.8V CMOS logic (74AUCXX) for re-clocking would be one option to allow a simple series resistor to form the sole signal conditioning circuit.

The high capacitance from the input transistor base to substrate (~12pF) causes fast edges to couple into the substrate, which forms a bulk resistance of tens of Ohm to -15V.

All NPN transistors in the IC sit on this substrate, so noise there is coupled back into the entire output DEM circuit. So it appears at the output.

At the same time, as all logic baring the input is differential all the way to the bit switches, the logic is pretty much unaffected by the noise in the substrate.

Only the leakage of the edges into the audio output are of real concern.
"Stopped clock" operation eliminates BCK related noise (great idea) at the moment of conversion (when it matters most), but does not eliminate other contaminants.

The above is an urban legend. There is no "moment of conversion" where removing the mainly BCK and Data related noise matters.

It is likely still an improvement, as typically stopped clock uses a higher frequency for BCK than needed, causing other potential issues, but being easier to filter.

And the period during which this easier to filter noise exists is only 50% or less of the time. So any process using averaging (e.g. FFT) will show improvements.

If no signal conditioning is used and especially if the source is modern CMOS logic with very high edge rates, stopped clock can provide objective and likely subjective improvements, but it is preferred to use correct signal conditioning instead.

Thor
 
  • Like
Reactions: HenSch
The above is an urban legend.
From page 8 of the PMD100 datasheet.
Stopped_Clock.png
 
I can't understand what I see on the picture. X=L, Y=R, there is some phase difference (20 us or so), but why? Freq 440 Hz, I/V is DC coupled.
The conversion does not happen at the same time in both channels? The difference seems in the order of 1 WS time.
 

Attachments

  • 20241012_130309.jpg
    20241012_130309.jpg
    461.6 KB · Views: 52
I can't understand what I see on the picture. X=L, Y=R, there is some phase difference (20 us or so), but why? Freq 440 Hz, I/V is DC coupled.
The conversion does not happen at the same time in both channels? The difference seems in the order of 1 WS time.

How do you determine if there is a time difference if the X axis is no longer time?
I'm not that experienced with X-Y mode. But I think you are misinterpreting this.
 
No. Data of both left and right should be latched when BCK goes up after WS has gone down.
According to the I2S bus specification (see attached), R channel is latched when WS is high, and L channel is latched when WS is low, on the next leading edge of the SCK. That means 1/2 WS period, but I measured a full WS period between the channels.
If there is time difference between the channels, be it 1 WS period or 1/2, simultaneous mode looks better.
In my test pin 27 is connected to pin 28, this should select time multiplexed 2s complement format.
 

Attachments