Hi Marcel, I can’t seem to get my head around this concept. I thought U8/U15 are like dc servos to null the outputs. So, how do they function as BAL to SE converter like a typical differential Amp?
Hi Bohrok,
What sound card are you using.
I’m amazed what dynamic range you seem to be able to cover.
In your image I can see a 262.144 point FFT, but what is the sampling frequency, 44.1 Khz ?
That would mean a filter bin width of ca. 0.17 Hz.
To compare your noise level to 1Hz bins, one would have to add 7.7dB to the noise level . One Herz is the bin width I’m always using for easier calculations.
in your case a Blackman-Harris window increases the spectral content per bin.
A rectangular windows does not, but maybe your software takes this into account ?
A last question, what is the noise you measure when connecting the output of your soundcard to its input in a loop test ?
Hans
What sound card are you using.
I’m amazed what dynamic range you seem to be able to cover.
In your image I can see a 262.144 point FFT, but what is the sampling frequency, 44.1 Khz ?
That would mean a filter bin width of ca. 0.17 Hz.
To compare your noise level to 1Hz bins, one would have to add 7.7dB to the noise level . One Herz is the bin width I’m always using for easier calculations.
in your case a Blackman-Harris window increases the spectral content per bin.
A rectangular windows does not, but maybe your software takes this into account ?
A last question, what is the noise you measure when connecting the output of your soundcard to its input in a loop test ?
Hans
@Hans Polak: Thank you for your feedback in #959. However, what I was really after was the DSD rates you used for your distortion and IMD figures on pages 1 & 2. I have been experimenting some with NoDACs recently (the sound may be outstanding IMHO) and while they/I may achieve quite impressive results at e.g. DSD128 (THD at ~-104 dB or higher and a very clean spectrum) the distortion figures I get rise very sharply with DSD256 and DSD512.
As far as I can see you do not mention the DSD rate(s) in relation to these measurements (or then I have missed it)?
@bohrok2610: Thank you also for following up and posting your results above. Hmmm... much lower distortion level changes 🤔 ... I wonder if my AD7760 EVM board may be fooled by the high level of ultra-sonic noise when just using an RC filter ... Anyway, something to consider - & your ES9038Q2M result is indeed impressive
BTW a bit of experiences with HQPlayer (which I have been using for my measurements): On the EC5 & EC7 modulators the noise floor actually drops up to about 55 kHz (DSD128, a little lower for DSD64) after which it rises "sharply". I reckon Jussi from Signalyst somehow noise shapes the response to be like this but if one intends to use HQPlayer for playback it could allow for a higher filtering cut-off frequency.
Cheers, Jesper
As far as I can see you do not mention the DSD rate(s) in relation to these measurements (or then I have missed it)?
@bohrok2610: Thank you also for following up and posting your results above. Hmmm... much lower distortion level changes 🤔 ... I wonder if my AD7760 EVM board may be fooled by the high level of ultra-sonic noise when just using an RC filter ... Anyway, something to consider - & your ES9038Q2M result is indeed impressive
BTW a bit of experiences with HQPlayer (which I have been using for my measurements): On the EC5 & EC7 modulators the noise floor actually drops up to about 55 kHz (DSD128, a little lower for DSD64) after which it rises "sharply". I reckon Jussi from Signalyst somehow noise shapes the response to be like this but if one intends to use HQPlayer for playback it could allow for a higher filtering cut-off frequency.
Cheers, Jesper
No sound card. DAC is DUT and ADC is AK5394 as seen on https://www.diyaudio.com/community/attachments/measurement-setup-jpg.1199175/.
So all my DAC measurements are in a sense loop tests.
For noise measurements you should use V/rtHz scale as then the bin width is fixed. E.g. as in https://www.diyaudio.com/community/attachments/lt3045-15v_noise-jpg.1199182/. But for measuring SNR most software are able to calculate it properly in dBFS/dBc/dBV scale regardless of FFT window or size.
For lowest noise floor I use PCM and ES9822PRO as ADC. Here is the PCM noisefloor of ES9038Q2M/ES9822PRO with bandwidth from 2Hz to 96kHz.
Unfortunately ES9822PRO suffers from "ESS IMD hump" and has high order HD artefacts which is why I don't use it for DAC THD measurements.
Last edited:
Hi Jesper,@Hans Polak: Thank you for your feedback in #959. However, what I was really after was the DSD rates you used for your distortion and IMD figures on pages 1 & 2. I have been experimenting some with NoDACs recently (the sound may be outstanding IMHO) and while they/I may achieve quite impressive results at e.g. DSD128 (THD at ~-104 dB or higher and a very clean spectrum) the distortion figures I get rise very sharply with DSD256 and DSD512.
As far as I can see you do not mention the DSD rate(s) in relation to these measurements (or then I have missed it)?
@bohrok2610: Thank you also for following up and posting your results above. Hmmm... much lower distortion level changes 🤔 ... I wonder if my AD7760 EVM board may be fooled by the high level of ultra-sonic noise when just using an RC filter ... Anyway, something to consider - & your ES9038Q2M result is indeed impressive
BTW a bit of experiences with HQPlayer (which I have been using for my measurements): On the EC5 & EC7 modulators the noise floor actually drops up to about 55 kHz (DSD128, a little lower for DSD64) after which it rises "sharply". I reckon Jussi from Signalyst somehow noise shapes the response to be like this but if one intends to use HQPlayer for playback it could allow for a higher filtering cut-off frequency.
Cheers, Jesper
The THD and IMD tests were performed both at DSD128.
Hans
Thanks. What I find remarkable about those results is that no THD compensation is needed. ES9038Q2M is set to SYNC mode (without ASRC). With PCM input 3rd HD is much higher so THD compensation is required to reach similar levels. Also ES9038Q2M in PCM mode has much more high order artefacts as shown here. These DSD results are somewhat strange as ES9038Q2M is supposed to convert DSD to PCM internally for volume control so it may be that PCM input causes higher RF glitches than DSD.
Thx for explaining, I missed that in your previous posting, my fault.
So you are using a 60dB 0.33nV/rtHz LNA and an ADC at some sample frequency displayed with a window that is all calculated in your software down to V/rtHz, Right ?
Sorry for my questions, reason is to trying to understand why you are measuring a much higher S/N than Marcel and I did.
What is the maximum signal your LNA can handle, because that's the reason I could not measure the S/N for DSD64 because of the huge HF content .
And is this LNA SE or Balanced, I tried to fid Scott's LNA but failed to find it, because in balanced form S/N will be most likely -3dB lower.
Your ADC most likely has a sharp filter to prevent aliasing from the large HF content in your spectrum up to 20Khz, right?
Hans
Last edited:
Look at the feedback cap on the common mode opamp. The time constant is too short to only affect average DC offset. It will work to correct dynamic AC offset/imbalance too (do common mode noise attenuation). IIUC the circuit, that's essentially what balanced to SE conversion does.
Last edited:
Yes, for the 2 noise measurements I used a 60dB LNA. The software I use is either REW or Multitone Analyzer neither of which is my software.
LNA is SE. More about it here. My version has 4 parallel 2SK369s which bring the noise density to about 580pV/rtHz. I also have Samuel Groner's LNA which goes to about 400pV/rtHz.
SNR measurements were made without LNA. So normal loopback from DAC to ADC in balanced mode.
Yes, AK5394 has a filter but I don't think that is the reason for no HF content with DSD64. My guess is that the SW you used for DSD generation is responsible for the HF content.
Yes, AK5394 has a filter but I don't think that is the reason for no HF content with DSD64. My guess is that the SW you used for DSD generation is responsible for the HF content.
It's not the software I use, it's the spectrum of DSD after filtering on Marcel's board.
See image below.
It's exactly this huge HF content that oversteers my LNA, making it impossible to measure S/N.
Hans
It's not the software I use, it's the spectrum of DSD after filtering on Marcel's board.
See image below.
It's exactly this huge HF content that oversteers my LNA, making it impossible to measure S/N.
Hans
Attachments
Last edited:
But does it also ‘sum’ the inputs like typical differential Amp? Any other reference for this circuit about how this BAL-SE is accomplished?
We sometimes call it 'differential summing.' Its taking the difference of two signals, one of which is inverted relative to the other. Doing that attenuates common mode noise. Doesn't matter the exact topology used to do it, just the end result that matters.
Also, maybe consider what it would be doing if it were a DC servo. That would just mean it attenuates only very low frequency common mode noise. At least, you could think of it in those terms.
No. As I said SNR measurements were made in DAC-to-ADC loopback in balanced mode.
I also tested with Marcel's trick (continuous 01010101... pattern). The resulting noisefloor (see below) is more or less the same as ADC's noisefloor. So nothing wrong with the analog part or the measurement setup. Problem is most probably in the digital circuit. I need to test that more.
Ok, thanks!
In that case all good to go with SE only by just taking half of the final output stage - knowing that the balanced output from the DAC core gets converted fully into unbalanced line output with good CMRR. So do we still need to worry about tidying up other matters as thought before?
IME, yes, we are probably going to keep finding ways to make it sound better for awhile. Offhand, there are three areas I would like to check.
First, looks like it should be possible to drive it with the AX-22 clocks and using Andrea's FIFO board. I will need to install the u.fl connectors on Marcel's dac board to try it. Also, expecting some slightly longer u.fl cables early next week.
Second, the power supply design for the switched output resistor arrays is not even a Jung regulator. The sound of those particular op amps to some degree is probably being convolved with the analog audio output. It may turn out that discrete shunt regulation would end up sounding better, not that I'm offering to implement that. Again, this is IMHO and IME. Don't know if it will make much or any difference with PSS measurements.
Third, output stage uses a number of IC op amps. Again, PSS measurements may look fine, but complex music signals are not PSS.
Why bother? Marcel's dac continued to improve in terms of subjective SQ overnight. But the improvements are getting smaller and smaller. Right now Macel's dac has some good promise, but its not good enough yet that I can compare his basic FIRDAC to Andrea's FIRDAC. I could compare the two dacs as they are, of course, but it would be plainly obvious to anyone how different they sound and which would they would take if they could afford it. That's the truth. I know some people won't like it and won't want to believe it. Marcel is deservedly a local hero here for his knowledge and his generous contributions to diy. That doesn't mean this particular FIRDAC is as developed and refined now as some of the best dacs in the world. Its good alright, but not that good. Not yet anyway.
Some examples:
- The dynamics of music waveforms are flat, compressed sounding.
- Instruments are still missing low level details, and sort of mixed/blurred together without the space that should be between them. Still a slight bit grainy/distorted/veiled (whatever you want to call it) which may be masking some details.
- Imaging is unfocused, blurry.
Last edited:
@Markw4 ... I look forward to hearing your & your visitors' impressions. BTW which output stage are you using together with Andrea's DSD DAC? You also mention an AX-22 clock: searching the internet doesn't really give any result - might you have a link to a description of it?
@bohrok2610:
I am not into the specifics of the ES9038Q2M but looking at the functional block diagram at the beginning of the datasheet it does look, well, a bit surprising. However, I admittedly am not that much into the specifics of this so I can't really comment any further ... But outstanding DSD result, IMHO.
And, so, now it is weekend - best wishes for your weekends ;-)
Jesper
@bohrok2610:
I am not into the specifics of the ES9038Q2M but looking at the functional block diagram at the beginning of the datasheet it does look, well, a bit surprising. However, I admittedly am not that much into the specifics of this so I can't really comment any further ... But outstanding DSD result, IMHO.
And, so, now it is weekend - best wishes for your weekends ;-)
Jesper