My dongle search rabbit hole

[markanini]

CX31993 is basically transparent to me, very similar to my AK4376 powered Fiio BTR1. Samsung dongle sounds like a light lo-fi effect is mixed in with the highs. Note that these are my subjective impressions, not making any claims or anything, probably the tuning of the IEM/Headphone used will affect one's perceptions, among other things.

 

[OK1]

The CX31993 is not "screechy" or lacking in bass. It has it's annoyances but it's not audibly bad.

I was reporting what I heard on the "copy" I received. I have only heard one CX31993 dongle, an it was really easy to hear this variance, compared to the Apple dongle and the TempoTec Sonata BHD.

Have obviously not heard every CX31993 in existence, but mine definitely sounds like the observations, I posted earlier.

The issue is - there are so many CX31993 products in existence, from so many different suppliers. So many. Maybe if I had not been such a cheapskate, I'd have settled for either the Abigail, or one of the CX based JCally dongles.

The CX I have now does what I need it to do - enable voice calls and listen to conferences, so the audio quality is NOT critical. And I did not spend much money on it.

For critical listening, I am 100 satisfied with my Sonata BHD, will most likely never be replaced for another 5 to 10 years, as my primary dongle. Just hope it it remains reliable, as it has been so far. Not a single glitch, and it has ASIO drivers. Tempted to buy another one as a spare. Cos I am so happy with it and also with the cost. Can be bought for about $35 or less. Definitely fantastic value for money, rivalling every other high quality dongle DAC and exceeding the threshold for audible DAC quality. i.e in theory, if there is a better measuring DAC, it is most likely diminishing returns, i.e better, but unlikely to be audibly better.

 

[OK1]

CX31993 is basically transparent to me, very similar to my AK4376 powered Fiio BTR1. Samsung dongle sounds like a light lo-fi effect is mixed in with the highs. Note that these are my subjective impressions, not making any claims or anything, probably the tuning of the IEM/Headphone used will affect one's perceptions, among other things.

The problem with all these budget devices, is at first they seem to put out the good quality stuff, so the reviewers can publish nice comments about them, then who knows, they then proceed to silently version the product, to most likely save money.

I was not expecting anything out of this world. Actually did not know what to expect, cos the published specs of the CX seem pretty decent, but what I heard is definitely lacking compared to the Apple dongle and the BHD. The BHD is audibly more precise and detailed than the Apple dongle.

One interesting difference the CX has. Inserting headphones comes with pops and clicks in my ear. Not so with the Apple and BHD dongles, which are silent when inserting or disconnecting headphones. No pops clicks, static in my IEM. The CX also has some kind of short static broadband noise on insertion and deletion, really short, like half a second, and then silence. No constant background noise on the CX, just this brief whoosh, on insert or disconnect of the headphone, in the headphone socket of the DAC.

The advantage of the CX and the BHD, are the higher sample rates. Not that I need higher sample sample rates with the CX, cos of my use case for these (speech on voice calls). I've tried "upsampling" all audio I'm listening to on the BHD, to 96K, with the intention that whatever DAC filters are in use, are less aggressive. Of course this could be all subjective, cos I have no way of measuring. 96K listening on the BHD, seems to add something - or rather remove something compared to running it at 48k.

I like the 96K better than the 48K, on the BHD, at this time. A little more clarity on the top end(like introducing a little bit of the crispiness I heard on the CX). Only time will tell, if this opinion remains so.

 

[mc.god]

I got the same GraveAudio DA06 for €5.50 along with Sonata BHD and honestly can't hear a difference, nor with JM20 or Ugreen Hifi Pro. My hearing memory is simply too short to rely on for making any comparison when I have to stop playing, change the dongle where iem is connected and set that dongle as default audio output in the OS, and even if I happen to hear a difference I know it's most likely in my imagination knowing that they are different sources.

 

[Basic Channel]

I was reporting what I heard on the "copy" I received. I have only heard one CX31993 dongle, an it was really easy to hear this variance, compared to the Apple dongle and the TempoTec Sonata BHD.

Have obviously not heard every CX31993 in existence, but mine definitely sounds like the observations, I posted earlier.

If you got the graveaudio one that was cheap as chips recently from graveaudio themselves (as part of the buy 3 and get free shipping thing) then I have the same one, and imo it's audibly fine. It's got a bit of a dodgy connection for me.

I am not disputing Aliexpress has complete nonsense (i bought a 16tb USB flash drive) and admittedly you might have younger and more refined ears or listening abilities than me. @mc.god has the same dongle from graveaudio and mentioned the JM20 which I'd get if it was 10 EUR or so, but I'd be astonished if I could hear a difference.

I suppose we all have our thing, I am currently buying millions of eartips from Aliexpress. All the eartips are mine. More eartips than braincells over here.

 

[OK1]

The ASIO on the BHD crashes, needing an occasional restart of my DAW, every once in a while, but I am a pretty heavy user, with extreme CPU and RAM use near to the limits of what's possible on my computers. I found this unusual, cos on my other ASIO devices, I would run for as much as a whole month, sometimes as long as two months maybe more, without restarting the computer and nary a glitch on ASIO based outputs - this was on my EMU 0404 USB audio interface.

So I'm running WASAPI on the BHD, to avoid these occasional crashes. Have not had any crashes so far on WASAPI exclusive. Yes I like the 96K upsampling on the BHD, just that extra crispness and tightness in the upper frequencies. But this could be placebo, or confirmation bias. But I like the "air" and slight crispiness this adds. More lifelike. if a bit more digital. But I feel I'm hearing more of what the DAC is capable of. My subjective opinion.

One possible explanation is lack of cramping from my EQ plugin, when running at 96K. Yes the bundled parametric EQ in Reaper, my DAW cramps at 48K, but not at 96K. combined with less aggressive DAC filtering @ 96K. The BHD, to the best of my knowledge, does not have any user adjustable filters.

 

[mc.god]

The ASIO on the BHD crashes, needing an occasional restart of my DAW, every once in a while, but I am a pretty heavy user, with extreme CPU and RAM use near to the limits of what's possible on my computers. I found this unusual, cos on my other ASIO devices, I would run for as much as a whole month, sometimes as long as two months maybe more, without restarting the computer and nary a glitch on ASIO based outputs - this was on my EMU 0404 USB audio interface.

I'm using all my dongles on Linux at native sample rates of the source files, tested just once the Sonata BHD asio driver on windows to confirm functionality but nothing more.

The BHD, to the best of my knowledge, does not have any user adjustable filters.

You are right, our BHD doesn't have adjustable filters, it's a feature of the BHD Pro with its fpga usb bridge.

 

[Basic Channel]

One possible explanation is lack of cramping from my EQ plugin, when running at 96K. Yes the bundled parametric EQ in Reaper, my DAW cramps at 48K, but not at 96K. combined with less aggressive DAC filtering @ 96K. The BHD, to the best of my knowledge, does not have any user adjustable filters.

From (bad) memory, cramping is more of an issue when you are doing crazy boosts up high like "air EQ".

Actually here is the video from the smooth-voiced Mr Worrall

 

[Nutul]

(...) I would run for as much as a whole month, sometimes as long as two months maybe more, without restarting the computer (...)

You are on a Windows box, I suggest you DO restart your computer at least once a week.

 

[OK1]

If you got the graveaudio one that was cheap as chips recently from graveaudio themselves (as part of the buy 3 and get free shipping thing) then I have the same one, and imo it's audibly fine. It's got a bit of a dodgy connection for me.

I am not disputing Aliexpress has complete nonsense (i bought a 16tb USB flash drive) and admittedly you might have younger and more refined ears or listening abilities than me. @mc.god has the same dongle from graveaudio and mentioned the JM20 which I'd get if it was 10 EUR or so, but I'd be astonished if I could hear a difference.

I suppose we all have our thing, I am currently buying millions of eartips from Aliexpress. All the eartips are mine. More eartips than braincells over here.

I'm listening A/B between the CX and the BHD, both at 96K. Here is what I think.

From listening, the mids on the CX are definitely more pronounced, and a bit more harsh - just a bit. The BHD's sound more balanced and less fatiguing than the CX, more natural. So the BHD mids like human voice, actually seems a bit recessed compared to the CX. i.e I have to raise the volume on the BHD's to hear the mids as clearly.

The highest frequencies on the CX seem relaxed, while on the BHD's these are more pronounced.

MY EXPLANATION.

When one reads the specs of these devices, there is a tolerance, in these dongles, of up to 0.5 dB deviation, in the case of the BHD. Can't recall what the deviation is on the CX. Compared to some professional audio devices, where the frequency deviation is typically no more than 0.2 dB across the audio spectrum, and in more recent times, the best professional devices, can achieve as low as 0.1 dB deviation, or in some cases much lower deviations.

I find that this is one area where consumer dongle DACs vary from the higher quality DACS - adherence to very exacting frequency response targets.

So consumer DACs beat budget professional DACs, in measurements like SINAD and S/N and Dynamic Range, but even budget professional audio interfaces tend to hit no more than 0.1 dB deviation in frequency response.

I will add, in many consumer dongle DAC's, they omit the FR deviation in their specs.

In comparison - here are the line out specs for a prosumer/professional USB audio interface, the Audient ID4 - note they use THD+N @ -1 dBFS, which is a bit of "cheating", somewhat lenient standard of measurement, to make their device measure better (my opinion).

The BHD's specs for comparison, are better in most parameters, except Frequency Response

So the consumer world and the professional audio - budget devices, seem to have different focusses. Consumer audio highlights dynamic range and SINAD. Professional audio is more stringent in their Frequency Response.

I think this is laxity in FR, is what introduces the possibility for such a difference between two dongle DACs the TempoTec Sonata BHD and the CX31990, which I am able to hear.

One more interesting issue. There is no FR specification for the CX31990 - I have been searching for the official specs, never found any. And none of the products specs for any DAC Dongle based on the CX31990 that I have seen, omits specs for the Frequency response. Typically the only reason specs are ever omitted, is if there is something to hide. My personal suspicion is that the FR specs of the CX31990 based devices leaves a little to be desired, and this explains what I'm hearing. I would think the FR specs of the CX are worse than the BHD, if measured, and this would not surprise me.

Of course all this is hypothetical, just me trying to explain, what my ears are telling me.

 

[markanini]

I'm inclined to a smilar explanation for perceived differences. Something that amounts to a small-magnitude broad tilt over the audible frequency band. Occams razor.

 

[OK1]

You are on a Windows box, I suggest you DO restart your computer at least once a week.

I've regularly run Windows laptops for upwards of 3 months without restarting, just put them to sleep when the lid was closed, nothing more. I come from a computing/information technology background where I was directly responsible for configuring Windows servers and deploying them into mission critical with zero downtime for months (or years), and I'd ask them to reboot the servers no more than once every year or so - to recover from possible memory leaks - just in case., which is where I got into the habit of running Windows operating systems, for long periods without restarting.

 

[Nutul]

I was directly responsible for configuring Windows servers and deploying them into mission critical with zero downtime

Then you know your way with Windows, no need for further advice. I just wouldn't use it for audio, AS-IS. You most probably know how to get the best out of it, no comment.

 

[OK1]

I'm inclined to a smilar explanation for perceived differences. Something that amounts to a small-magnitude broad tilt over the audible frequency band. Occams razor.

I'm glad for one thing. At least the measurements give us some hint as to the root causes, of what we are hearing.

From (bad) memory, cramping is more of an issue when you are doing crazy boosts up high like "air EQ".

Actually here is the video from the smooth-voiced Mr Worrall

My assertion that this may be due to cramping or rather lack of cramping, was just that - an assertion, I have not done any significant A/B comparing just this factor alone. But I could, by using two parametric EQ's one which cramps and another which does not (because it has oversampling implemented internally). I could if I wanted to, but life's too short. Am enjoying the 96K listening, whatever the real reason why it sounds better.

I had been struggling to hear Lewis Capaldi's vocals, which was pretty strange, cos over several weeks of listening to all kinds of music, his was the only one where I struggled to hear what he's saying (but in a way not surprised cos he's Scottish !!- they speak a different English language - pretty hard to hear what they say sometimes). But 96K does improve the intelligibility.

Then you know your way with Windows, no need for further advice. I just wouldn't use it for audio, AS-IS. You most probably know how to get the best out of it, no comment.

Yeah I do my best to do all the usual optimisations for Windows. Albeit Windows 10 is relatively stable, even without these optimisations.

What I'd say with Windows is - avoid putting a lot of non audio critical software on an audio centric computer. For something like a professional studio, i.e install the barest minimum of apps. And this applies to all the manufacturer bloatware, avoid where possible, by doing a clean install from the ISO that comes directly from Microsoft, and keep device drivers and BIOS, up to date. Less is more. If possible even things like Microsoft Office, avoid installing - most of those documents can be read by Google apps. Lean and mean machine, in my opinion, is best for optimal Windows stability and performance.

Of course be careful where you visit on the web, and think before you click.

 

[markanini]

I don't worry too much about filter cramping myself, most of my projects are 24/44.1 But that's because I rely on plugins that use internal oversampling. This setup leads to lower required HDD data transfer rates so I can use more silent HDDs, and disk space needs are also reduced.

 

[Basic Channel]

My assertion that this may be due to cramping or rather lack of cramping, was just that - an assertion, I have not done any significant A/B comparing just this factor alone. But I could, by using two parametric EQ's one which cramps and another which does not (because it has oversampling implemented internally). I could if I wanted to, but life's too short. Am enjoying the 96K listening, whatever the real reason why it sounds better.

I had been struggling to hear Lewis Capaldi's vocals, which was pretty strange, cos over several weeks of listening to all kinds of music, his was the only one where I struggled to hear what he's saying (but in a way not surprised cos he's Scottish !!- they speak a different English language - pretty hard to hear what they say sometimes). But 96K does improve the intelligibility.

I am Scottish and I can understand exactly what I am saying, ken woat a mean?

96k shouldn't make vocals more intelligible, even with cramping which I think only becomes a thing with large amounts of EQ. You could hear the difference by doing a null rather than an A/B. The standard hearing test (which is for speech intelligibility) tests from 150-7.5k.

The simplest solution is to have a setup that refuses to play Lewis Capaldi's music

 

[OK1]

I am Scottish and I can understand exactly what I am saying, ken woat a mean?

96k shouldn't make vocals more intelligible, even with cramping which I think only becomes a thing with large amounts of EQ. You could hear the difference by doing a null rather than an A/B. The standard hearing test (which is for speech intelligibility) tests from 150-7.5k.

The simplest solution is to have a setup that refuses to play Lewis Capaldi's music

Glad you could poke a bit of jesting @ the Scots. I wish I could live there, extremely unique beautiful landscape which I see on the TV and documentaries, never been. Must visit soon. Tons of serene wide open spaces.

Yeah - my hypothesis on cramping, was just that, any cramping introduces @ 48K, change from the tiny amount of shelving I use on top, which is a cut of only 1dB @ Nyquist centered at 8.3K, has negligible cramping. I do have an alternative EQ that does NOT cramp, cos it oversamples internally, but parametric EQ's are not good friends, cos it's difficult to apply the same parameters from one EQ to another, cos their implementation of Q (Bandwidth per octave), in my case, is not identical. Would be too much trouble to bother to try to match the EQ precisely in my non-cramping EQ.

It's a moot point, cos @ 96K, this cramping moves to 96K, and the effect of any cramping @ 96K, is inaudible, in the audible frequency spectrum.

I think the improvement I'm hearing is far more to do with the relaxed DAC filtering @ 96K, compared to the filters applied @ 48K.

 

[Basic Channel]

Glad you could poke a bit of jesting @ the Scots. I wish I could live there, extremely unique beautiful landscape which I see on the TV and documentaries, never been. Must visit soon. Tons of serene wide open spaces.

Well I am from Leith in Edinburgh so the landscape is more 'trainspotting' than 'Weir's way'

I also moved away 20 years ago to an even lesser populated place!

I hope this comes across as intended (honest and not rude) but id urge you to maybe listen again for these differences you are describing and maybe consciously bias (or question) yourself the other way.

 

[OK1]

I don't worry too much about filter cramping myself, most of my projects are 24/44.1 But that's because I rely on plugins that use internal oversampling. This setup leads to lower required HDD data transfer rates so I can use more silent HDDs, and disk space needs are also reduced.

I did an experiment a few years ago, to remaster a stereo track. Went all the way to 300+ Khz sampling rate. I sincerely cannot remember what equalizers or compressors I used on that project, cos it is a while ago. It was done via a virtual audio interface that comes with Reaper, known as Dummy.

So there is absolutely no need to output any audio. Most budget audio interfaces such as the EMU 0404 USB that was my daily driver audio interface, top out at 192K, so even in offline mixing mode in the DAW, I'm not sure how high I could raise the sample rate, with the real device still selected, as the output - have not really tried anything higher than 192K in offline mode with a physical audio device.

Anyway the result was pretty startling. After downsampling to 48K, the top end became "sharper" but glossy, so much easier to hear the effects of compression. The challenge is - not all plugins play nicely above 192K, so one has to either use a trusted plugin from a developer like Waves, which does a good job of publishing what sample rates their plugins support, or actually test each plugin individually, using something like PluginDoctor, or in my case a bundle of analyzer tools running in a dedicated DAW project - where I carry out such tests, to see or myself, what kind of distortion is being created by plugins, and what sort of noise if any is inherent in a plugin.

My thinking, from a purist perspective, is in theory, if I could run all plugins at really high sample rates, like 700K, during mixing, and for compressors during mastering, @ even higher sample rates, to completely avoid the need for oversampling, or at least reduce the amount of internal oversampling needed in compressor plugins, the final result would more closely approximate what occurs in analog gear. Whether that is good or bad, is a separate consideration.

Why? It may be negligible, but with every plugin which oversamples internally, which is the correct implementation when running @ 48K or 44.1K, I still feel that something is lost, and furthermore we are at the mercy of the quality of that oversampling implementation. I have not done comparative experiments, cos even if there was a benefit, my current gear which I use regularly, would grind to a halt, if I regularly wished to use higher sample rates above 48K, sometimes I'm already struggling @ 48K and need to bounce things down, to fit everything in the CPU available. So for me the effort to do a lot of experimentation, at this time is pointless, cos I cannot take any advantage of the improvement at very high sample rates, in my DAW, in real time cos of CPU limitations. Typically I'm mixing over 70 tracks in a song, with tons of effects.

But I do have a compromise. I mix at 48K(or 44.1K depending on what my client sends me) which I'll refer to as "single", most of the time, but when done with mixing, the final stereo result, bounced down i.e mixed offline @ 96K (or 88.2K), so lets call this "double". Whenever I compare the double to the single, the double always sounds better. Always. I could postulate all manner of reasons for this, in my case, I am sure some of the bundled plugins in Reaper are compromised. For example the Reaper parametric EQ cramps, at single sample rates, and their default compressor does not oversample. I have been using these tools for over a decade, partly cos they generally get the job done, and at single sample rate, have very low CPU utilisation, so in spite of their well known and well documented flaws (huge thanks to Dan Worral), its hard to get away from them. This may explain why my offline double mix always sounds better than the single. I do use other EQ's which do not cramp, and compressors with oversampling, but these two staple plugins are used all over my mixes, hard to get away from them.

Going forward, I'm more likely to be mixing @ 96K, when I have a CPU that can handle this, so I'm listening the mix, during mixing, at the desired target rate.

The theories related to oversampling are great, and it is wonderful to have a good reliable valid explanation for what we are hearing, but at the end of the day, we all now have the liberty and choice, to try out various options and listen to the end result. Then decide.

Something I'm looking forward to, whenever I am brave enough to upgrade to version 7 of Reaper, is the selective oversampling, where one can decide how much oversampling each plugin will achieve - from the DAW, and Reaper is responsible for the up and downsampling, so this is done in a uniform manner, rather than relying on the quality of oversampling chosen by each plugin's developer. That will be an interesting experiment, so for EQ's I could choose dual i.e 2x, and for any dynamics processors (compressors, limiters, etc) - maybe 8 x. and this is all done in real time.

What I am especially pleased about, is we have reasonably priced dongle DAC's like the TempTec Sonata BHD which for less than $40, level the playing field, and almost anyone can now listen to good quality audio, with fewer compromises.

 

[OK1]

Well I am from Leith in Edinburgh so the landscape is more 'trainspotting' than 'Weir's way'

I also moved away 20 years ago to an even lesser populated place!

I hope this comes across as intended (honest and not rude) but id urge you to maybe listen again for these differences you are describing and maybe consciously bias (or question) yourself the other way.

Trust me. I have listened over and over again, comparing the CX and the BHD, and the difference is exactly as I have described. I was fully expecting that there would be absolutely no difference between them, looking at their Signal to noise, Dynamic Range, and SINAD(THD+N). I expected that the audio of the CX would be above the threshold, where any audible difference would be heard. Where they differ measurably in their quoted specs, and significantly is in their SINAD. But what I'm hearing is a mid boost or more distortion in the mid frequencies, such as the vocals on the CX, and a more relaxed mid range/vocal on the BHD.

Taking into consideration the most likely higher variance in frequency response (cos it is not quoted anywhere - which is a red flag), as well as the variance in SINAD, this easily explains what I'm hearing. I fully went into this expecting NOT to hear a difference, and it's only because I cannot deny what I'm hearing and this difference surprised me, that I'm bothering to share my experience here.

My take is this, modern recent IEM's have reached a level of resolution, that it is becoming easier to hear these differences, which some may assert, should not be audible. The real take for me is that :

1. Clearly there is a correlation between the measured results and what I'm hearing - and this cuts across the CX, BHD and the Apple dongle. What I hear is consistent with the measured stats.

2. In any future purchase decision, beyond Signal to Noise, Dynamic Range, SINAD, I'll also be taking into consideration, frequency response.

3. I'm raising my bar on these measurements, for devices I would want to buy in the future, to something like :

3.1 Signal to Noise >=120dB
3.2 Dynamic Range >= 120dB
3.3 SINAD (-120dB)
3.4 Frequency response no more than +/- 0.1 dB, from 20 to 20K. (or any other measurement which predicts how accurate the frequency response can be expected to be)

That's about where, based on what I have bought so far and what my ears have heard, one can expect that, anything better, may not be audible, or any improvement is negligible, and we have gone into diminishing returns territory.

I really like Lewis Capaldi @ 96K on the BHD. Now I have to go back and listen to all my favourite tracks @ 96K on the BHDs, with the same IEM (CCA CRA - 2023 revision) and eartips - the large ones that came with the IEM.

The BHD only falls short, compared to my revised target, in the SINAD (slightly) and in the Frequency response spec. But I am happy with it. Very happy with it value for money - great, very good value for money. I do think this is the level 2 to aspire to, with the Apple dongle being a level 1, the starting point in acquisitions, cos its cost vs quality is a no brainer., in the search for better.