I don't know but can be the gates & flip-flops integrated in the chip are better than external flip-flops or the added flip-flops generate more jitter due to PCB traces, etc.
Can be a lot of reasons the answer.
But in this case I agree KIS is the way to go.
The 4th order passive filter designed by abraxalito (thank you) using Janzen wax coils & FT-3 teflon russian capacitors avoid any kind of pop or click to the sound, only adding a bleeder resistor at the output to discharge the output capacitor that can be eard only one time when you start your listening session.
Can be a lot of reasons the answer.
But in this case I agree KIS is the way to go.
The 4th order passive filter designed by abraxalito (thank you) using Janzen wax coils & FT-3 teflon russian capacitors avoid any kind of pop or click to the sound, only adding a bleeder resistor at the output to discharge the output capacitor that can be eard only one time when you start your listening session.
Last edited:
It'd be great to get rid of those too but without them the signal wouldnt exist.
you get some CMR with the output's flip flop balanced signal but marlins words give the impression it takes away from the ''effect''.
If inside modern DAC chips like Sabre and AKM most internal processing will be avoided with a native DSD input, a similarly transparent sound would be expected.
DSD and PCM from my Sabre DAC sound very similar and nothing like the transparency of DSD from the XMOS, I even prefer PCM over DSD with the DAC.
If the suggested circuits can preserve most of the XMOS sound that will be fantastic, some degradation is unavoidable but the Sabre may be much worse in comparison.
If you don't want to add anything at all to the signal path, not even increase the number of flip-flops from 10000 to 10002, the only possibilities I see to reduce the noise are:
1. Improve the supply of the XMOS board somehow. The supply is the reference for the DAC, any high-frequency ripple on it will mix out-of-band quantization noise back to low frequencies. If you can somehow reduce the ripple by better decoupling and/or regulation, the noise floor should get lower.
2. Replace the XMOS board with an Amanero board. No idea what this will do.
3. Experiment with different R and C values for the filter. For a given time constant, the higher the resistance, the less the filter will be affected by differences between the output resistance of the XMOS board for high and low output levels.
4. Use the lowest available DSD bitrate, as you had already established that this helps.
1. Improve the supply of the XMOS board somehow. The supply is the reference for the DAC, any high-frequency ripple on it will mix out-of-band quantization noise back to low frequencies. If you can somehow reduce the ripple by better decoupling and/or regulation, the noise floor should get lower.
2. Replace the XMOS board with an Amanero board. No idea what this will do.
3. Experiment with different R and C values for the filter. For a given time constant, the higher the resistance, the less the filter will be affected by differences between the output resistance of the XMOS board for high and low output levels.
4. Use the lowest available DSD bitrate, as you had already established that this helps.
Or use an AKM dac in direct DSD mode.. obviously, in the right way, right circuit, topology, attention to clocks, power, all. I have heard it in a direct comperison to a DSC1 V2.6 . Although the the dsc1 did sound really (and surprisingly) good, in my (biased) opinion the AK ( modded Mirand) dac was still better. More refined presentation.
And there is a strong difference between PCM and DSD256 in dsd direct, native through that ak4490 based dac. Even if pcm is driven by the best filters from HQP, 384k/24 bit.
And indeed while the DSC1 sounded good, it occasionally did burst into noise ridden sequencies, (some slight high pitch whistle, and we did hear some suspicious noise artifacts between songs too.
The real dac based no-dac configuration is just dead silent, very low distortion, highly refined sounding, explosive dynamics and no clicks when switching rates and modes..
And indeed while the DSC1 sounded good, it occasionally did burst into noise ridden sequencies, (some slight high pitch whistle, and we did hear some suspicious noise artifacts between songs too.
The real dac based no-dac configuration is just dead silent, very low distortion, highly refined sounding, explosive dynamics and no clicks when switching rates and modes..
Last edited:
thank you for the report, the direct DSD mode of AKM is beyond what is possible with Sabre then? the block diagram for AKM show it is very direct indeed, Sabre doesnt look to have much special attention to DSD in their diagrams.
Thanks for suggestions,
right now the power supply is dual regulated 3.3V with LT3045, however the DIYinHK XMOS notes the addition of DC-DC converter on board for 1.2V supply on the newer revision of their XMOS board, due to impractical power requirements (800ma vs 300ma) with 1.2V LPS. this could be contributing to this problem.
Will try changing the adjusting the filter too, actually when more RC stages were added to filter I thought signal to noise level did change but was hard to tell with treble rolloff
Exactly MvdG,
The problem is usually implementation, which at these speeds is not trivial.
The big advantage of the RTZ format is it should also cancel other noise
to some extent.
I would think that fully balanced with some form of RTZ would be the best
way to go. There is also another variation of RTZ which I will try and dig up
that is more efficient.
It would have been great if something like HQplayer had this as an OP
option.
T
increasing resistance of filter made a slight difference but at the same time quality changed for worse, doing opposite and lowering resistance (22ohm, 0.33uf) improved SQ further, noise did get slightly worse.
DSD128 sounds best right now, a steeper filter might be necessary for DSD64. DSD256 sounds worse and the noise is unbearable.
There was 2k pot after the filter, it could have been controlled better using HQPlayer digital attenuator but noise/pops come out unattenuated so it wasnt an option
Will leave the DC-DC converter for now as its not going help the uneven rise and fall times, should help in the future when main noise issue is fixed.
Anyway the Sabre DAC is definitely a terrible example to compare to RTZ, there are even 2 op amps in the signal path after the chip.
DIYinHK have AK4493 kit for $40, which would be easy option.
DSD128 sounds best right now, a steeper filter might be necessary for DSD64. DSD256 sounds worse and the noise is unbearable.
There was 2k pot after the filter, it could have been controlled better using HQPlayer digital attenuator but noise/pops come out unattenuated so it wasnt an option
Will leave the DC-DC converter for now as its not going help the uneven rise and fall times, should help in the future when main noise issue is fixed.
Anyway the Sabre DAC is definitely a terrible example to compare to RTZ, there are even 2 op amps in the signal path after the chip.
DIYinHK have AK4493 kit for $40, which would be easy option.
Last edited:
Or when they called opa1688, or 1656 or 2156..
Diyinhk kit: I do have an older version. It's not bad but presently not comparable..
And needs a lot of attention in the execution.
I did not follow it lately, is the software mode activated recently? Previously one had to organize an external controller. Also look out for the dsd-direct mode correctly implemented..
And I did not find yet a trustable comparison between 4490 / 4493, when the former is driven from ~7V or similar high VREF. It cannot be done on 4493.
What I'm trying to hint is the actual realization is all important..
Ciao, george
I have seen Yanasoft assembling a dsd no-dac as well..
With a slightly different approach, separate logic gates
Diyinhk kit: I do have an older version. It's not bad but presently not comparable..
And needs a lot of attention in the execution.
I did not follow it lately, is the software mode activated recently? Previously one had to organize an external controller. Also look out for the dsd-direct mode correctly implemented..
And I did not find yet a trustable comparison between 4490 / 4493, when the former is driven from ~7V or similar high VREF. It cannot be done on 4493.
What I'm trying to hint is the actual realization is all important..
Ciao, george
I have seen Yanasoft assembling a dsd no-dac as well..
With a slightly different approach, separate logic gates
Last edited:
Noise from no dac's dsd dacs are usually the known hf noise that is a result of the noise shaping and the result of inherent low bit depth.
The HF noise is not a big problem per se.
Most noise that gets overlooked is the ISI, Inter Symbol Interference, which is a result of the limitation of the rise and fall times of the digital pulses. Besides being limited, the fall time can be different than the rise time. The ISI can lead to modulation like chirping sounds, birdy tones and just plain noise or harmonic distortion products and is easily (and best) measured at around -40 to -60 dB. Just use 1KHz as a signal and see what the noise floor as well as harmonic distortion products do..) This also explains why many classical music as well as the beginning or end of music exhibit these noises and/or tones.
For practical hardware implementation, any real life implementation needs analog filtering, but this means that rf-design needs to be taken into account. I'm betting many of the experiments suffer from rf reflections that change dramatically when a different size/brand/type or value of capacitor/resistor/coil is used. I know I made those circuits;-)
Together with its jitter sensitivity and low psrr it can be a real surprise what's actually causing what kind of artefact and it would nice to see HQPlayer make some kind of rtz scheme. It is really easy to make with some ic's though (And gate and flip flops), but signal goes down 6 dB and the hf noise actually is higher in level. Another way is to interleave 2 dacs, but it gets complicated this way really easy and kind of unnecessary for audio dacs. A good pdf is from tektronix: google "isi dac tektronix" and it should be on top.
In short: you need really high rise and fall rates, or use RTZ encoding, HQPlayer's AMSDM7 works way better than the other modulators, though I wish there was a rtz modulator in it
The HF noise is not a big problem per se.
Most noise that gets overlooked is the ISI, Inter Symbol Interference, which is a result of the limitation of the rise and fall times of the digital pulses. Besides being limited, the fall time can be different than the rise time. The ISI can lead to modulation like chirping sounds, birdy tones and just plain noise or harmonic distortion products and is easily (and best) measured at around -40 to -60 dB. Just use 1KHz as a signal and see what the noise floor as well as harmonic distortion products do..) This also explains why many classical music as well as the beginning or end of music exhibit these noises and/or tones.
For practical hardware implementation, any real life implementation needs analog filtering, but this means that rf-design needs to be taken into account. I'm betting many of the experiments suffer from rf reflections that change dramatically when a different size/brand/type or value of capacitor/resistor/coil is used. I know I made those circuits;-)
Together with its jitter sensitivity and low psrr it can be a real surprise what's actually causing what kind of artefact and it would nice to see HQPlayer make some kind of rtz scheme. It is really easy to make with some ic's though (And gate and flip flops), but signal goes down 6 dB and the hf noise actually is higher in level. Another way is to interleave 2 dacs, but it gets complicated this way really easy and kind of unnecessary for audio dacs. A good pdf is from tektronix: google "isi dac tektronix" and it should be on top.
In short: you need really high rise and fall rates, or use RTZ encoding, HQPlayer's AMSDM7 works way better than the other modulators, though I wish there was a rtz modulator in it
my experiment on no DAC
As long as I have tested, the best logic which has the same rise and fall time is an LVDS receiver like ADN4664.
https://www.analog.com/media/en/technical-documentation/data-sheets/ADN4664.pdf
The propagation delay is excellent ,ie.tpdhl=2.15ns,tpdlh=2.03ns(typ). ADN4664 must have clean power since it is 0dB PSRR. The best driver for LVDS is FPGA because it has built-in LVDS driver and has the ability to adjust propagation delay between tdhl and tdlh. Internal PLL can digitally compensate for the difference by 30ps step(spartan 6). You can also use it for DSM. So, Ideal design requires two devices, an FPGA with IIS(PCM) input and an LVDS receiver for DSD(1bit DSM) output. I haven't implemented that way, but it had probably -90db SNR(almost no audible noise).
As long as I have tested, the best logic which has the same rise and fall time is an LVDS receiver like ADN4664.
https://www.analog.com/media/en/technical-documentation/data-sheets/ADN4664.pdf
The propagation delay is excellent ,ie.tpdhl=2.15ns,tpdlh=2.03ns(typ). ADN4664 must have clean power since it is 0dB PSRR. The best driver for LVDS is FPGA because it has built-in LVDS driver and has the ability to adjust propagation delay between tdhl and tdlh. Internal PLL can digitally compensate for the difference by 30ps step(spartan 6). You can also use it for DSM. So, Ideal design requires two devices, an FPGA with IIS(PCM) input and an LVDS receiver for DSD(1bit DSM) output. I haven't implemented that way, but it had probably -90db SNR(almost no audible noise).