Mark, I remember a private discussion you and me had, a few months back. This is an excellent moment to try not only changing the primary phase of the power supply transformer, but even more so the secondary phase of the windings. If you do, do not look for better tonality, but for the more or less engaging sound and do check every used secondary winding you have.
Marco
Marco
Marcel, it should be the case that RTZ does mitigate the noise floor going up when the sampling rates go up, right? Do you still think crosstalk is the main culprit and do you see low hanging fruit regarding minimizing it in this dac?
From my personal experience, these types of dacs can all sound very effortlessly when those level of distortions and 16 bits are achieved and they are such a good base to try various ideas and philosophies on. Since it`s so basic and modular it is very easy to tinker with, from layout, to component based, from cabling to type of psu, even logic families, types of filters, solid state to tube can be changed, as well as how to feed the beast from a pc etc. Last but not least, is what happens here: a bunch of guys exchanging knowledge, interests to create progress and solve issues. I think this is what keeps me interested in diy audio: sitting thinking, tinkering and creating on your own, then having a place to share and evaluate with like minded people.
Thanks Marcel, for having succeeded in making another great project!
Sorry for my enthusiasm, but I had to share this.
Marco
From my personal experience, these types of dacs can all sound very effortlessly when those level of distortions and 16 bits are achieved and they are such a good base to try various ideas and philosophies on. Since it`s so basic and modular it is very easy to tinker with, from layout, to component based, from cabling to type of psu, even logic families, types of filters, solid state to tube can be changed, as well as how to feed the beast from a pc etc. Last but not least, is what happens here: a bunch of guys exchanging knowledge, interests to create progress and solve issues. I think this is what keeps me interested in diy audio: sitting thinking, tinkering and creating on your own, then having a place to share and evaluate with like minded people.
Thanks Marcel, for having succeeded in making another great project!
Sorry for my enthusiasm, but I had to share this.
Marco
Have they? You are the third person posting measurements in this thread, after me and Hans. Thanks for posting!
Cool, you are the third person designing a variant of the DAC, after Thorp's PCM version and nautibuoy's simplified filter board layout. In fact I hoped people would use it as a starting point to be redesigned as they like, nice to see that that is indeed happening.
Here is another picture of the RTZ dac next to my ES9038Q2M dual dac. Both use same form factor for input and output connectors which makes it easier to swap dacs in chassis.
View attachment 1199170
Here is a picture of the measurement setup. ADC is AK5394. DAC and ADC are powered from my SilentSwitcher which in turn is powered from a power bank. My USB-I2S board uses USB power. It can be powered from a 5V battery but I have not found any reason to do that neither in listening nor in measurements. Please note the high-end audiophile power bank from Lidl.
View attachment 1199175
Ok. Then the measurements. I used Multitone Analyzer as it has built in DSD measurement signals. Measurements were made with DoP64.
First 1kHz sine.
View attachment 1199176
Very good. Distortion is low and high order HD is missing. Noisefloor could be better. More on that later.
Just for comparison here is the same measurement with my ES9038Q2M DAC.
View attachment 1199177
It is worth mentioning that with DSD input ES9038Q2M THD is much cleaner than with PCM as all the high order harmonics are below the noisefloor. But noise is a bit higher than with PCM input.
Ok, back to RTZ.
Here is 60Hz sine:
View attachment 1199178
As with 1kHz high order HD is below the noisefloor.
CCIF 19k/20k:
View attachment 1199179
Very good.
Let's get back to the noise performance.
Here is SNR (i.e. DAC playing silence):
View attachment 1199180
So SNR(A) is -97.5dB. I'm not sure if this can be directly compared to Hans' measurements as I have typically seen slightly worse noise numbers with Multitone Analyzer.
The noise floor is considerably higher than what Hans and I measured. Of course it's a different board layout, but I don't know if that has anything to do with it.
I'm sure the LT3042 is at least as complicated, but it's all hidden away in a small black package. A hard to hand solder small black package: small pitch and exposed die pad.
Anyway, my main reason for designing a variant of the 1964 Hilbiber bandgap reference is that I found it more fun than just using an LTsomething. If you notice any performance advantage, it will be at very low frequencies, where C8 doesn't yet filter the reference noise much. (As you know, low-frequency noise at the reference translates to close-in amplitude noise around the desired signal.) At higher frequencies, it are mainly the OPA2210's determining the reference noise, rather than the actual bandgap.
May appear a bit intense but does uncover interesting details and ideas as we go along.
In regards to SE/BAL, I see that your own setup is practically SE due to the fact that your HeadPhone Amp used for tests and serious listening is Unbalanced type - only using one leg of the DAC output. . So, you are not using the full ‘benefits’ of a Balanced DAC. Next, this DAC does not have transformer output compared with ValveDAC, DSC2+. So improvements in sound quality as you have experienced could be due to less distortion from a well designed electronic output stage.
FYI, the original Signalyst DSC1 is SE with electronic output stage. Then someone came along and made a balanced version and then found that to drive the next stage, most likely SE gear, easiest to plug in a transformer. Andrea’s discrete DAC has similar arrangement.
Only when it goes up due to differences between rising and falling delays.
I do still think that because of my experience with a ribbon cable that had to be split, and the differences in noise floor between Hans and me. It is very well possible that I'm barking up the wrong tree, though.
Regarding low hanging fruit: use the U.FL connectors and don't place the 0 ohm resistors that connect the data and bit clock pins of the Amanero-compatible header. Then again, bohrok2610 uses U.FL, as far as I can see, and has the worst noise floor...
I'm glad you like it!
@Marcel, how to shut down the other half of your DAC for truly SE engine to suit?
Perhaps less switching noise this way, thank you
Last edited:
BTW, been letting Marcel's dac run. Its sounding better as I2SoverUSB clocks continue to stabilize after sitting for in storage for at least several months. I am reminded that previous experiments showed it took 3+ plus days of continuous run time on new clocks to fully audibly stabilize. Similar thing we noticed when Topping D90 was new. Took around 3 - 4 days for it as well.
You could remove R63 and R106 from the filter board if you like, the negative side of everything after the first filter stage is then silent. I don't think it has any practical advantage to do so, but it doesn't harm the single-ended output signal either.
The core of the DAC (I mean the part around U7, U9, U14 and U16 on the DAC board) needs to be differential to keep the current drawn from the reference as data-independent as possible. The first filter stage with its common-mode loop (U8, U15 on the filter board) is the differential to single-ended converter when you use it in single-ended mode.
My experience is that the main reason that you don’t get the the full potential of a balanced connection is because of pin1 connected to analog ground on the receiving side causing unwanted equalisation currents.
I wrote an article supported by Bob Whittlock in Jan’s Linear Audio called “One End Only” meaning that pin1 is only connected on the sending side but not on the receiving side.
https://www.linearaudio.net/pin-1-revisited-adding-oeo-accepted-rule
To transfer a signal you only need two wires and not three, a the third one in this case is causing unwanted disturbancies.
So a very simple test is to disconnect pin1 on your receiving amp and listen to the result.
It should improve the sound compared to SE or at worst be equal.
Hans
I wrote an article supported by Bob Whittlock in Jan’s Linear Audio called “One End Only” meaning that pin1 is only connected on the sending side but not on the receiving side.
https://www.linearaudio.net/pin-1-revisited-adding-oeo-accepted-rule
To transfer a signal you only need two wires and not three, a the third one in this case is causing unwanted disturbancies.
So a very simple test is to disconnect pin1 on your receiving amp and listen to the result.
It should improve the sound compared to SE or at worst be equal.
Hans
I think you need three, pin #1 to signal gnd on transmitting side (DAC). pin #1 not connected when tested my AKM DAC ->Benchmark BAL Amp and got totally distorted sound! I thought I saw this type of advice somewhere, that pin#1 is like a shield only… very surprised and wasted my time trying to troubleshoot the board.
Then followed as what data sheet indicated and all good afterwards.
Last edited:
No. I use MMCX connectors but that has nothing to do with the noise floor as you can see from my measurement of ES9038Q2M which also uses MMCX.
These noisefloor measurements are not directly comparable as the used test signals or the measurement setups were not the same. The higher noisefloor in my measurement has nothing to do with USB-I2S board as the same board with ES9038Q2M results in almost 15dB lower noise.
At least with DS dac chips for lowest noise the Vref circuit should be placed very close to the target with as low impedance as possible to the DAC Vref pins. LT3042 with its small footprint and small number of required external components makes this possible. My hunch is that also in this RTZ dac LT3042 would lower the noise floor.
So why do you only need two wires with SE.
Balanced with three wires are two SE signals fighting with each other.
Condition however is that the two balanced wires have a CM potential that’s within the CM range of your receiver.
If not you will get very distorted sound like you obviously experienced.
The solution is simply that the chassis of both sender and reciever are both connected to mains ground.
Hans
P.s. when experimenting with PCB’s not yet mounted in a chassis, connect the mains ground to the individual sender and receiver ananolog grounds via a 10R resistor.
Last edited:
Here are 1kHz measurements at DoP128. Measurement setup is the same. My USB-I2S board can do DSD256 but only in Linux as I have not implemented an ASIO driver. DoP256 would require higher clocks on my board.
RTZ 1kHz DoP128:
Strangely noise is 5dB worse. I'll need to study this further to see if something is amiss on my RTZ board.
To rule out the measurements signal or UBS-I2S I also measured ES9038Q2M with DoP128 as reference.
In this case noise is almost 3dB lower with DoP128. So nothing wrong with measurement signal or USB-I2S.