From a modicum of listening over the last day comparing two dedicated to each sample rate DACs, both me and my wife preferred the 2XOS version. The margin for me was quite small, 2XOS being a little more engaging, more difficult to stop listening - my wife says the 2XOS version seems to have more variation in loud/soft than the NOS. Guess audiophiles might call that 'better macrodynamics'. This suggests the next generation of PhiDAC is probably going to include a digital filter of some kind, in the meantime Foobar users can employ an upsampler like SoX.
Incidentally PCB mods for rev2 are well underway, revised schematic to be posted in the next couple of days.
2xOS will be cleaner, airy, sparkless, faster sounding and wider soundstage, on the expense of lower texturing and a real natural sound. Your preference will be dependent on the kind of music you are listening. For the acoustic instruments in classic and jazz, some rock I prefer NOS. It gives me more musical reproduction with trusted transients and harmonics. A modern music (hipop, trance, most of pop) is mastered with attention to the dynamics and effects, with disregard to the natural reproduction, as it is known that most of people will listen it on DS converters where these natural vibrations are lost. It also takes time to train our brain how to focus on music and true harmonics. It doesn't come immediately. It is why many people prefer OS.
How many times OS? It depends on a jitter. My Audio GD R2R11 has virtually no jitter and majority of people say that 8xOS sounds the best.
In my testings differences are very low, I can hear only on headphones and is probably subjective. But I do upsampling 24-bits wide, as my DAC support it, here it will be 16-bit, it will produce more distortions.
It is worth to note that upsampling in Foobar with SoX is much better quality than any microcontroller can do, FPGA is expensive. You will have to add a bigger headroom for intersample peaks (IP), losing few dB of a dynamic range, so a typical increase in SNR will be not so big in the end. In a properly configured Foobar you can use replay gain feature to avoing distortions linked to IP's, but I think it doesn't work properly with SoX at the moment.
How many times OS? It depends on a jitter. My Audio GD R2R11 has virtually no jitter and majority of people say that 8xOS sounds the best.
In my testings differences are very low, I can hear only on headphones and is probably subjective. But I do upsampling 24-bits wide, as my DAC support it, here it will be 16-bit, it will produce more distortions.
It is worth to note that upsampling in Foobar with SoX is much better quality than any microcontroller can do, FPGA is expensive. You will have to add a bigger headroom for intersample peaks (IP), losing few dB of a dynamic range, so a typical increase in SNR will be not so big in the end. In a properly configured Foobar you can use replay gain feature to avoing distortions linked to IP's, but I think it doesn't work properly with SoX at the moment.
That's your prediction for how my DAC will sound based on the sound of your DAC? Its wrong for my DAC - there are crucial implementation differences between yours and mine - in particular the passive filter and the droop EQ - which may explain why its wrong.
How many MIPs would you say SoX is using to do 2XOS? I rather suspect you're not completely up to speed with what microcontrollers can do nowadays but happy to be proved wrong on that.
@abraxalito. What makes you so challenging? You did actually follow my recommendation by abandoning these multiple tiny shielded sound polluting coils, didn't you? I only came with a good faith, this is a proof. I can be wrong regarding super powers of MPU you have, but it shouldn't make you sick or depressed.
If you're really looking to be my therapist you're welcome but its way, way OT for this thread. Open another in the Lounge if that's your thing.
Back to DACs - here's the latest schematic for the rev1 PCB which will hopefully go out to manufacture today. The major change is there are now footprints for six paralleled DACs rather than four. I saw available real estate on the PCB and figured it would better be put to good use. I haven't yet modified the component values to accept six DACs though so for the time being the schematic is still effectively for a 'PhiDAC Quad' with two no-fit DAC sites. To run with 6 DACs needs a lowered supply voltage to keep the peak output current the same - fortunately that's just resistor value tweaks.
Back to DACs - here's the latest schematic for the rev1 PCB which will hopefully go out to manufacture today. The major change is there are now footprints for six paralleled DACs rather than four. I saw available real estate on the PCB and figured it would better be put to good use. I haven't yet modified the component values to accept six DACs though so for the time being the schematic is still effectively for a 'PhiDAC Quad' with two no-fit DAC sites. To run with 6 DACs needs a lowered supply voltage to keep the peak output current the same - fortunately that's just resistor value tweaks.
I am not your terapist, I do contribute to this project the same as others do.
You have turned around 180 degrees and trashed these multiple SMD coils. This move was in a direct result of my contribution to this thread and now with a proper filter the project looks even more attractive. However it surprises me very much how you do express your appreciation.
I'm not so sure that this is accurate. Back in September last year you were making significant contributions - for which I am happy to show appreciation. Your question about adding more chips got me thinking which is the kind of contribution I like to encourage. However more recently your postings have been less than constructive, even downright delusional on occasions. I've not seen that level of disingenuity from any of the other contributors to this thread(*) and I remarked on a particularly dishonest post at the time.
If you want to blow your own trumpet about influencing my choice of inductors, be my guest but not on this thread. I won't be joining you in your self-congratulation.
Back to DACs - the PCB has gone out to manufacture now and here's a snapshot - in red this time around 🙂
(*) on second thoughts, @Hollowman is an exception.
Attachments
Last edited:
I know it does not really matter, but I do love the symmetry on the board. Very nice layout.
I do have a component question though. Does the selection of the BC856B matter much? I see that the average is 100-200mw and 100mhz. I found this one BC856B-13-F Diodes Incorporated | Mouser
that has double those values. If I were selecting custom components would this have any benefit, would it have no influence or would it be detrimental?
I do have a component question though. Does the selection of the BC856B matter much? I see that the average is 100-200mw and 100mhz. I found this one BC856B-13-F Diodes Incorporated | Mouser
that has double those values. If I were selecting custom components would this have any benefit, would it have no influence or would it be detrimental?
The transistors aren't very critical in terms of gain selection (that's the A/B/C after the number). I specified B because that's what I have to hand here but A and C would work equally well. I just checked Mouser and the C grade is cheapest by a whisker so by all means use that one. BC856, BC857 or BC858 will all work fine as they only differ on breakdown voltage and all have more than enough voltage capability to work at 10V.
BC857CLT1G ON Semiconductor | Mouser
BC857CLT1G ON Semiconductor | Mouser
I use 2XOS in one system (as preferred), and no oversampling in another, mainly because I don't trust going to 4XOS, as the only simplistic option, and I just haven't tried 4XOS yet.
An appealing technical reason for using 2XOS is in consideration of the Nyquist. For a 44.1kHz sampling rate this requires rejection at 22.05kHz. For 20kHz bandwidth this requires filtering over 1/10th octave (brick wall filtering). At 88.2kHz requires filtering at 44.1kHz, hence over 1 octave compared to 1/10th octave. At 176.4kHz this is over 2 octaves. Hence benefits of 4XOS are minimalistic in the comparison to that of 2XOS. It seems that 4XOS and above can do more harm than good above 2XOS.
All of this can be moot for those who believe that extensive filtering is over-rated on the D/A side, being fundamentally applicable on the A/D side.
I had a brief listen at 4XOS to my PhiDAC Quad, didn't notice anything different from 2XOS but it was such a short listen I'd not draw any conclusions from it.
The way I look at it there's 4.1kHz (from 20k to 24k1) for the filter's response to fall from passband to stopband. I use different numbers from you because I assume there's no content above 20kHz on the recording. Which might not be a very secure assumption in practice. So I get 2/10ths of an octave. Still very, very steep.
48k1 vs 4k1 for the transition band from my numbers so almost 12X easier for the filter.
I get 136k vs 48k so roughly 3X easier still or 33X easier than NOS. Its worth pointing out that in terms of digital filter complexity its much easier to go from 2XOS to 4XOS than it is from NOS to 2XOS. That's because going from 2X to 4X the digital filter doesn't need to be very steep and just as with analog filters, a steep digital filter is more complex (more taps) than a gentle one.
I've seen arguments about trying to avoid ringing on the DAC end without realization its already included on the recording by virtue of the anti-aliasing filter on the ADC.😛
This may be OT, but I found this iFi whitepaper unique in how they balance all the key tradeoffs with filters (image rejection, frequency response, and especially transient response).
Based on their data about the tremendous sensitivity of the human to timing anomalies, they conclude that "44.1kHz sample rate digital audio lacks time domain resolution to be transparent," but they have settled on a compromise that they believe gets close enough...
Based on their data about the tremendous sensitivity of the human to timing anomalies, they conclude that "44.1kHz sample rate digital audio lacks time domain resolution to be transparent," but they have settled on a compromise that they believe gets close enough...
It is interesting. The sensitivity of human's to timing anomalies, being tremendous or otherwise, is dependant upon using equipment to conduct real world testing. This is the highly questionable part in legitimizing claims as to absoluteness in suggesting "tremendous" performance improvements in accuracy and transparency. In being called a "whitepaper" it eludes to a "purity in truth" normally reserved for legitimate scientific research. Seemingly this paper is limited to results suggested proven by "MQA" researchers, not by those necessarily unbiased. Nevertheless, this is not to suggest that their compromise isn't a legitimate one that could have "tremendous" impact sonically.
Digital manipulation can have extraordinary influence on sound reproduction, yet the extent of such influences does not appear equally applicable to all devices in a DAC conversion system. Sufficiently advanced analog networks, ones that can be highly transparent at 44.1kHz, can preclude the need for more extreme digital filtering that can "cause" lesser analog networks to behave more transparently.
I fully agree with, including a note above about MQA. On the other side there is observed a big commercial success of Chojo Mojo/Hugo2/Qutest/Dave range. These devices have a digital filter with an extraordinary number of taps, sort of super DS converter. Marketting? I decided to follow a trend for a modern compensated R2R design like Audio GD, Denafrips or Airist. Basic models are quite affordable starting from $350.
Coming back to the project. I am very happy that the latest incarnation of a LingDAC receives a proper analog network. It is in result of my contribution, while not acknowledged by a project leader, it cannot be denied. As you wrote, it is true, a good analog filter is sufficient, there is no need for the undepowered DSP oversampler/filter. Besides, a dispute NOS vs. OS slowly becomes irrelevant, as most of my current purchases are coming in hi-res format. 24/96 mastering is becoming a standard offering on many classic/jazz websites. Who is going to upsample 96kHz files? It doesn't make sense.
In other words, NOS is going to be mandatory for R2R implementations. I made recently some comparison of upsampled in Foobar with SoX 44.1kHz files and playing on Audio GD R2R11. Final conclusion is not yet. If you want to contribute, here is a double CD I suggest for testing: Ahmad Jamal, Yusef Lateef - Ahmad Jamal Featuring Yusef Lateef/ Live At The Olympia June 27.2012 (2014, CD) | Discogs
Last edited:
Its lacking context - the signals that are used to provide the 'impulse responses' of various filters aren't signals that can be encountered in normal situations. By which I mean they can't get recorded onto a format like CD. So the whole discussion about ringing might be moot if it can't ever happen in practice.
Before a signal gets onto a recording it needs low-pass filtering to avoid aliasing. That low-pass filter prevents frequencies above 20kHz getting in, since if they did they'd alias and that effect sounds particularly unpleasant. When considering 'ringing' of filters its necessary to investigate whether that 'ringing' can in practice be triggered by signals on a recording (as opposed to test impulses which can't ever get onto a recording).
To use an analogy from power amplifier testing - they're often tested with 10kHz squarewaves to see if there's any 'ringing'. But a 10kHz squarewave won't get recorded onto a CD seeing as to make the wave 'square' needs harmonics at 30k, 50k, 70k and so on, ad infinitum. If the squarewave is sufficiently low-pass filtered, no ringing is visible.
Following that logic, it will likewise be impossible to encounter a perfect square wave on a CD, since a square wave requires an infinite number of harmonics, and our recordings cut off above 20kHz.
That being said, NOS DAC designers seem to like showing scope images of square waves playing through their DACs, and those square waves do tend to look more correct than those I've seen coming out of DS designs. And I think they show these waveforms because they believe they correlate with sound quality via high performance in the time domain.
Yes, indeed, though its possible to get a reasonable approximation to one at a low enough frequency. Say 100Hz.
Yes, I've seen those too. But they're faked in the sense that they create those in some waveform editor and not by recording a squarewave from a signal generator through an ADC. And the fakeness is compounded by the lack of reconstruction filtering in most NOS DACs so that the sharp edges come through unscathed.
Quite likely, yep.
It is actually a good test also for Delta Sigma converters. It help discusing various filter design response characteristics and evaluate whether filter is tuned properly. It is not true that ringing is trigered by a pulse (which do not exist in any recording). People who make such claim did not attend university. There is no trigger mechanism, ringing is always there, proportionally to the dv/dt. Ringing is altering microdynamics all the time. Only a detection is triggered in our brain, when not heard as a ringing, it is heard other way.
As for discusion of the recorded bandwith, what do you think, should a brick wall 20kHz bandwith limit (with strong pre-recorded ringing) be also enforced for the higher sample rates? It is interesting to know your answer, as most of current recordings is made using completely differend technology and distribution of music is also at higher sample rates. When I play 24/96kHz file, I don't need a sharp digital filters to remove images, a much simpler analog filter do a job.
I am currently running a NOS TDA1541a DAC (inside an inherited Naim CD3). This is connected to a non-feedback I/V design capable of operating at input currents above +/- 10mA. This also has a uniform input impedance of around 3 Ohms into the MHz region. The 3dB rolloff of the I/V was limited to 1.5 MHz from around 5MHz for reasons that my tube preamplifier began reminding me of Sparky.
Theoretically there doesn't seem anything necessarily problematic. Digital recordings generate "line" sample data expected periodically reproduced. In the case of a current DAC, theory suggests that if current outputs are an exact proportion of the file value being generated instantly and is then held constant over an exact time interval, and then instantly changes to the next proportionate value, etc., then all audio data below 20KHz is preserved. It is considered that no problems arise from doing so if such perfection is maintained.
The rectangular nature of the waveforms, being squares waves or otherwise rectangular, seems can only have 3rd, 5th, 7th harmonics, etc., as being above 20KHz, otherwise they would reflect into the audio band below 20KHz, much like an AM detector. 2nd harmonics, as intrinsically coupled with a DC offset component, manifests as creating frequencies coming off from DC into the audio band from the modulation. Hence to maintain only audio below 20KHz requires such linearity in both the DAC and I/V with emphasis on maintaining symmetrical rectangular reconstruction.
To engage in reproducing all frequencies into the MHz region is not recommended as a finished product. Yet to do so can clarify an understanding of variant factors in the reproduction. Particularly in gaining incite into the degree linearity and filtering that is necessary. This is to suggest that if the audio reproduction is formidable with a 6dB/pole located at 1.5MHz, what benefit exists in removing the visible signs of the steps. What added value is expected using a brick wall filter or perhaps 2XOS? Filtering can obscure better performance. One question comes to mind: How faithfully can an I/V converter reproduce a step function waveform before filtering is applied?