HWZ Forums

Login Register FAQ Mark Forums Read

Creative Super X-Fi Amp measurements / RMAA / review

Like Tree21Likes
Reply
 
LinkBack Thread Tools
Old 05-11-2018, 08:11 PM   #1
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Creative Super X-Fi Amp measurements / RMAA / review

Part 0: Obligatory photo



From left to right: Super X-Fi, Ganaha Hibiki (CV: Numakura Manami), GTX 1060 screaming "You took me out of the PC to install a WHAT?", unused NS50 vouchers
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"
wwenze is online now   Reply With Quote
Old 05-11-2018, 08:11 PM   #2
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Part 1: Analog performance

Outputting at ~1.90Vrms

Unloaded
20Hz THD=0.0016% THD+N=0.029%
1kHz THD=0.00021% THD+N=0.00063%
10kHz THD=0.00083% THD+N=0.0011%

33 ohm resistive load on both channels
20Hz THD=0.0019% THD+N=0.029%
1kHz THD=0.00090% THD+N=0.0012%
10kHz THD=0.0014% THD+N=0.0017%

Output impedance
Around ~0.4 ohm @ 1.9Vrms 33 ohm, limited by Arta reporting down to only 0.1dB precision

I'll just take the result as less than 1 ohm, I'm sure other reviewers with their own websites and stuff and their audio analyzers can get a more accurate number.

RMAA summary


The noise numbers look a bit weird, the spectrum shows some weirdness which does not happen with Arta. Comparing the Arta spectrums, it is around 5dB worse than my Xonar DX, so it should still be getting -105dB or better.

Of course, having the same noise level as a computer soundcard is not a good thing for an earphone amplifier, because computers are meant to be connected to speakers with volume controls, while earphones are sensitive and connected to the output of the, erm, earphone amplifier, which means noise levels need to be way lower. Somebody once said the more meaningful measurement of noise is in terms of volts. But what if this amp also has a digital volume control inside? That can change the amount of noise in terms of volts. Noise is not easy to quantify (hence why sometimes you can hear noise coming from a system that supposedly has -80dB or -100dB or better noise), and I'm feeling lazy. We shall see.

Results discussion
The numbers are unbelievable, but the graph checks out: A tiny bit of 3rd harmonic poking slightly above -120dB line.

One thing I noticed and remembered is that Arta and RMAA give slightly different numbers. And this is why I keep both measurements - Arta for the RMAA haters, and RMAA for comparison with other RMAA users.

I forgot I was supposed to use 16 ohm, not 33 ohm. But since I already did that might as well. I still have to collect max output power data anyway.

Back to the SXFi, unbelievable numbers. I didn't know my Xonar STX can measure this low, and because of this these numbers may actually be the limit of the measuring soundcard and not the SXFi.

The goal is to verify Creative's claim of THD+N=0.00036% IMD=0.0022% and crosstalk=-75dB and so far it seems to check out. I give an allowance of up to 6dB, if my numbers are worse than the claims (as is usually the case) it can be due to settings and equipment. Especially when THD+N of 0.0006% is the lowest that I have actually seen from Xonar STX self-loopback.

-75dB crosstalk without load is however pretty mediocre. Let's hope it is due to the cables/connectors/adapters, there is still chance for measurements. I have a feeling that this is the actual unloaded specs however. In which case it is really really bad, I mean because lots of cheap stuff can do at least this good with load.

I'm going to assume the SXFi should hit all the claimed specs. These are amazing numbers. The amp is amazing too, taking just a minor hit in performance from a 33 ohm load at slightly over 100mW in output power. LG has a smartphone that can hit the same number, but it doesn't have as much output power. This level of performance can win many desktop setups, and is packed into the tiniest package possible. Poor crosstalk requires further investigation, would be interesting to check if it is power supply related.

Update 7/11/18
Nope, crosstalk is this bad. And this setup using a Windows 10 tablet running on battery as the source for SXFi is the best setup out of the 3 tested - The other two being using a smartphone and my computer which also houses the Xonar STX and consequently having lots of common-mode noise. All measurement numbers - THD, noise, crosstalk, the Windows 10 tablet is better than the rest.

There is something interesting about the noise floor, or rather, the dynamic range. Well, normally SNR and dynamic range are around the same, but in some cases they can differ. And SXFi is one case.

Yellow is Honor 8 Pro's 3.5mm out at max volume

Spoiler!

The higher the amp volume, the higher the noise floor. Also, if your Android volume is low but the amp volume is still high, the noise floor remains high. So you should always keep the Android volume max while reducing the amp volume.

This happens when Android volume has been lowered somewhat but amp volume is still max:
Spoiler!

So, at typical listening volumes, the SXFi noise level is comparable to Honor 8 Pro. At higher volumes, the smartphone cannot reach that high so not a comparison. And it does still keep the THD+N performance I can't fault it.

But what I will say is, this noise level is typical; while not bad, it is also not good enough to spend money on if your goal is lower noise. It is comparable to smartphones, so if that is enough for you, then SXFi is enough for you. However, other amps using mechanical volume control can attenuate both the signal and the noise.

With most of the measurements finalized, I can give it a judgement proper:

Results discussion
- Distortion values are still out of this world.
- Noise is... not exactly what I expected from a USD$150 product. Acceptable, but can do better.
- Crosstalk is really, really, really, really, really bad. -75dB is probably the absolute best case scenario - No load, good power, good source. Drops to around -50dB best case with 33 ohm load regardless of output volume, and around -45dB with 16 ohm load. If your source is bad, these values can get even worse.
- Output power / behavior with load, the problem is output power depends on load impedance, and crosstalk also depends on load impedance, I cannot say that the amp is not affected by high output power. Also, Creative's output power ratings seem to be peak values as opposed to RMS values. Also, I can only get around 292mW with 16 ohm load before the amp starts becoming unstable. While this is still pretty respectable, I am conflicted on whether I should score against not performing up to rated specs.

- The problem is that despite the excellent distortion and reasonable noise performance, crosstalk is really, really horrible, and you're going to hear -75dB or -50dB way earlier than -100dB or -110dB. And we are dealing with a technology that requires careful cross-feeding of different channels' sounds at different phases, I believe this requires a crosstalk performance that is top-notch. If you don't need the SXFi, it is very hard to recommend this product as an amplifier, just due to the crosstalk alone.

Verdict (combining parts 1 to 4)
Between the flawed analog performance, flawed SXFi effect, and the unreliable performance (clipping or "static") appearing sporadically, it is *very hard* to recommend getting this.

If the analog performance was good, it would have been a good reason to get this amp even if you don't need the SXFi effect. But due to the crosstalk, it cannot be called good. Unless you don't mind living with high crosstalk for some reason.

If you are thinking of getting it for the SXFi effect, try HeSuVi with Out Of Your Head Genelec Studio virtualization. If you can tolerate that and you want something that is 2 to 3 times more aggressive, get the SXFi.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 08-11-2018 at 12:00 AM..
wwenze is online now   Reply With Quote
Old 05-11-2018, 08:12 PM   #3
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Part 2: Accuracy of ear mapping

So here's the plan:

Record:
My profile A
My profile B
My profile C
Other guy profile A (Or I'll call it profile O)

Diffmaker original with each of the 4 profiles
Get 4 files: Diff A, Diff B, Diff C, Diff O

Diffmaker Diff A with diff B
Diffmaker Diff A with diff C
Diffmaker Diff B with diff C
Diffmaker Diff O with A, B, and C

Compare results

BTW if you don't know what Diffmaker is:
https://www.diyaudio.com/forums/ever...t-effects.html

For "other guy", I'm going to use David Bowie's mugshot.

Spoiler!

Result
After some problems and lots of retries and changes to the plan which I am not going to write, I arrived at two files:

Diff4 minus Diff3.wav
and
Diff4 minus Diff David.wav

And... Diff4 minus Diff David.wav is much softer than Diff4 minus Diff3.wav

Hmm... this isn't supposed to happen

However here's a disclaimer: When comparing the original file and the recorded file (SXFi off), I noticed the amount of difference recorded was varying with time - It goes from nearly the loudness of a normal file, to nearly silent, and back to the original loudness again. I think there is something in the setup causing this long-term variance. Or I just suck at Diffmaker. In any case, the plan was a failure.
lxXXxl, behwatch and tkl23786 like this.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 06-11-2018 at 02:09 PM..
wwenze is online now   Reply With Quote
Old 05-11-2018, 08:12 PM   #4
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Part 2.5: More on mapping accuracy, and lots of frequency response charts

The reason I chose Diff4 and Diff3 above is because I already knew Diff4 and Diff3 are both accurate mappings of my ear.

So, while mapping, I already saw the things that can go wrong during mapping.

- Like not detecting the earlobe or detecting extra earlobe
Spoiler!
- The face can be mapped at various vertical angles, or the mapped face does not cover the whole face
- I think the ear can be mapped at various angles as well
- The best one was when it thought the side of my face is the ear

Results - Frequency response

Profile 2 vs Profile 3
This two profiles are definitely different
Spoiler!

Profile 4 vs Profile 3
Now these profiles are quite close to each other
Spoiler!

Note that due to me using white noise to measure frequency response, the result will fluctuate:

Profile 4 vs Profile 4
Spoiler!

Hence anything below 1dB is close enough

Profile 5 vs Profile 4
Spoiler!

So profile 5 and 4 are definitely accurate. Profile 3 is a little off. Profile 2 is wrong.

David Bowie vs Profile 3
Spoiler!

So David Bowie is different from me.

However, seeing how similar David Bowie's is compared to my Profile 2, this makes me wonder if this occurs when the mapping is bugged. (Because I used a photo for David Bowie.) Different normal people may result in smaller differences if their ears are similar, perhaps like the difference between my Profile 4/5 and Profile 3 (which might be inaccurate). But this shows that the software does detect big differences when there are big differences, and small differences when there are small differences, and apply the appropriate processing. So for the most part, I'm going to say a "Yes", the personalization does work. Exactly how precise? I am not going to measure my ears many times and then measure many different people's ears to find out the signal-to-noise ratio.

This also means I need to mention a PSA: Please make sure your mapping is done correctly, or you may get a result that is way, way off.

Frequency response of SXFi, "Unknown Headphone", and various earphone/headphone selection

So, when SXFi is off, the amp is completely flat as it should be. But when it is on, I get the below frequency response:

Spoiler!

I think this was taken with one of the bugged mappings. But as you can see, the SXFi effect itself has a much greater impact on the FR than the mapping and the head/earphone selection. (Note the different vertical dB scale.)

So, I compared "Unknown Headphone" with some other head/earphones for the lolz.

Spoiler!

Note that FR measurements of sound effects are mostly meaningless because sound effects are not linear-time-invariant, which means the FR can change with the type of sound being presented.

Or basically these measurements are just for lolz only.
lxXXxl, behwatch and tkl23786 like this.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 06-11-2018 at 04:14 PM..
wwenze is online now   Reply With Quote
Old 05-11-2018, 08:13 PM   #5
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Part 3: Stereo content

So I enabled SXFi

...

And disabled it

And enabled it again...

...

And disabled it again

I get that the goal is to reproduce speakers in a room. But if you're reproducing cheap speakers in a large hall, I'd rather you not.

A large part is related to that frequency response. (Keyword: Related, but not necessarily caused by. The FR itself may be the result of that other root cause.) The "cheap" sound. It is unfortunate that I don't have any SXFi-certified headphone / earphone. I only got, like, some obscure models from unknown brands like Alessandro MS-1 and Klipsch X11, I mean what has Klipsch even done before, like manufacturing loudspeakers since 1946?

But looking at the FR data above, the selected headphone / earphone only contributes to a small fraction of the FR, the bulk of it is still due to the SXFi effect. And testing different selected headphone / earphone confirms this - Selecting different headphones / earphones results in a noticeable and possibly important difference, but the effect is still small compared to the whole SXFi effect.

The sound also feels more distorted (as in the high THD kind of distorted) and details are lost. Definitely sounds more hollowy and reverb-y.

More importantly, this is what the SXFi effect does: Soundstage is compressed towards the center and moved forward. But not all of the sounds get this treatment. Some are spread around you including to the back.

The normal headphone sound sounds like sound hard-panned to the left and right, and in terms of front-back positioning it feels like right in the middle. The SXFi does take the sound out of this position and moves it more outside, some more forward and some more around. But still feels like it is coming from the headphones.

And I'm not sure if this can already be achieved by other competing effects, or if it can be done without resulting in such a big change to the sound. It does shift the sound forward which is relatively unique, most likely thanks to the included surround-sound technology. One thing I will say, and will say again in the 3D part of the review, is that the SXFi effect is very aggressive.

Here's the thing. I have earphone / headphone. I have speakers. My earphone / headphone have sounds that sound like my speakers, except they sound like they are coming from the sides of my head but otherwise they sound pretty much the same. And I believe most of you reading will agree with this statement.

So, why did the creators of SXFi feel like they need to change the sound *THIS* much? This is not "experience that’s indistinguishable from a dedicated set of surround speakers" or "high-end multi-speaker system in a professional studio", to be blunt I don't know wtf this is. Maybe an attempt at the feeling of magical immersion beyond the normal speakers sound? Then they should have written that instead of trying to simulate speakers, because it gives less things for people to pick at, and less things to fail.

Btw just for the lolz, if you apply SXFi effect on speakers, you get a soundstage that is even more compressed into the center and even further away from you forward. Logical, because the sound of speakers is already more center and forward, unless you are lucky enough to have a big studio with speakers far apart.

Also, you can recreate the headphone sound on speakers by putting your head right in between the speakers and pointing them at you. And even without pointing them at you (i.e. leaving the speakers pointing forward) the effect is already pretty close. (because speakers can be considered point sources, after ignoring that the frequency response certainly changes with direction, and baffle effects, and cabinet depth... and multi-driver interactions... but for the most part, speakers can be considered point sources)
lxXXxl, behwatch, LiLAsN and 1 others like this.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 06-11-2018 at 05:12 PM..
wwenze is online now   Reply With Quote
Old 06-11-2018, 06:14 PM   #6
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Part 4: 3D

A good analogy would be: "3D displays"
(More accurately, stereo displays, but most people call them 3D displays)

You know how 3D displays have solid science behind, but when you watch them they still feel weird? Like you have to know how to look at the screen then the 3D effect will work? Be it Nintendo 3DS display, or those double pictures that you have to cross-eye and overlap the image or something.

It's the same here.

Left and right is easy to differentiate, but front and back is going to be hard. So we look out for audio cues that suggest the sound is coming from the back. Like maybe more "round", "dark", "reverby", "hollowy" or something. This has been done in games. And it is being done here.

In other words, you need to know how to listen to virtualized surround sound in order to hear what is being done here.

But SXFi's effect is very aggressive, that's fairly unique. It also creates an image in front, this is also fairly unique. It also seems to do better with multichannel content (Less things to fake and more things to work with I guess), but it still causes a significant quality drop in sound quality. Again likely due to the high aggressiveness.

In the best case, I would call this sound an acquired taste. In the normal case, I would disable it. Because the cost to SQ is too great and not worth whatever benefit it brings. It is like taking a pair of awesome front stereo speakers, and cutting it up into 7.1 crap ones. (And it doesn't even produce real 7.1 but I'm just saying.) So it is beneficial for like 3.5% of the time, but when the movie or game is doing nothing and is just having smooth music and a person talking in front, you want good sound for that but you can't get it. How about 7.1 good speakers instead? I don't know how hard this would be, but this would be the holy grail.
behwatch and tkl23786 like this.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 06-11-2018 at 06:29 PM..
wwenze is online now   Reply With Quote
Old 06-11-2018, 06:52 PM   #7
High Supremacy Member
 
delceer's Avatar
 
Join Date: Sep 2016
Posts: 34,276
Part 0: Obligatory photo



From left to right: Super X-Fi, Ganaha Hibiki (CV: Numakura Manami), GTX 1060 screaming "You took me out of the PC to install a WHAT?", unused NS50 vouchers
Part 1: Analog performance

Outputting at ~1.90Vrms

Unloaded
20Hz THD=0.0016% THD+N=0.029%
1kHz THD=0.00021% THD+N=0.00063%
10kHz THD=0.00083% THD+N=0.0011%

33 ohm resistive load on both channels
20Hz THD=0.0019% THD+N=0.029%
1kHz THD=0.00090% THD+N=0.0012%
10kHz THD=0.0014% THD+N=0.0017%

Output impedance
Around ~0.4 ohm @ 1.9Vrms 33 ohm, limited by Arta reporting down to only 0.1dB precision

I'll just take the result as less than 1 ohm, I'm sure other reviewers with their own websites and stuff and their audio analyzers can get a more accurate number.

RMAA summary


The noise numbers look a bit weird, the spectrum shows some weirdness which does not happen with Arta. Comparing the Arta spectrums, it is around 5dB worse than my Xonar DX, so it should still be getting -105dB or better.

Of course, having the same noise level as a computer soundcard is not a good thing for an earphone amplifier, because computers are meant to be connected to speakers with volume controls, while earphones are sensitive and connected to the output of the, erm, earphone amplifier, which means noise levels need to be way lower. Somebody once said the more meaningful measurement of noise is in terms of volts. But what if this amp also has a digital volume control inside? That can change the amount of noise in terms of volts. Noise is not easy to quantify (hence why sometimes you can hear noise coming from a system that supposedly has -80dB or -100dB or better noise), and I'm feeling lazy. We shall see.

Results discussion
The numbers are unbelievable, but the graph checks out: A tiny bit of 3rd harmonic poking slightly above -120dB line.

One thing I noticed and remembered is that Arta and RMAA give slightly different numbers. And this is why I keep both measurements - Arta for the RMAA haters, and RMAA for comparison with other RMAA users.

I forgot I was supposed to use 16 ohm, not 33 ohm. But since I already did that might as well. I still have to collect max output power data anyway.

Back to the SXFi, unbelievable numbers. I didn't know my Xonar STX can measure this low, and because of this these numbers may actually be the limit of the measuring soundcard and not the SXFi.

The goal is to verify Creative's claim of THD+N=0.00036% IMD=0.0022% and crosstalk=-75dB and so far it seems to check out. I give an allowance of up to 6dB, if my numbers are worse than the claims (as is usually the case) it can be due to settings and equipment. Especially when THD+N of 0.0006% is the lowest that I have actually seen from Xonar STX self-loopback.

-75dB crosstalk without load is however pretty mediocre. Let's hope it is due to the cables/connectors/adapters, there is still chance for measurements. I have a feeling that this is the actual unloaded specs however. In which case it is really really bad, I mean because lots of cheap stuff can do at least this good with load.

I'm going to assume the SXFi should hit all the claimed specs. These are amazing numbers. The amp is amazing too, taking just a minor hit in performance from a 33 ohm load at slightly over 100mW in output power. LG has a smartphone that can hit the same number, but it doesn't have as much output power. This level of performance can win many desktop setups, and is packed into the tiniest package possible. Poor crosstalk requires further investigation, would be interesting to check if it is power supply related.
Part 2: Accuracy of ear mapping

So here's the plan:

Record:
My profile A
My profile B
My profile C
Other guy profile A (Or I'll call it profile O)

Diffmaker original with each of the 4 profiles
Get 4 files: Diff A, Diff B, Diff C, Diff O

Diffmaker Diff A with diff B
Diffmaker Diff A with diff C
Diffmaker Diff B with diff C
Diffmaker Diff O with A, B, and C

Compare results

BTW if you don't know what Diffmaker is:
https://www.diyaudio.com/forums/ever...t-effects.html

For "other guy", I'm going to use David Bowie's mugshot.

Spoiler!

Result
After some problems and lots of retries and changes to the plan which I am not going to write, I arrived at two files:

Diff4 minus Diff3.wav
and
Diff4 minus Diff David.wav

And... Diff4 minus Diff David.wav is much softer than Diff4 minus Diff3.wav

Hmm... this isn't supposed to happen

However here's a disclaimer: When comparing the original file and the recorded file (SXFi off), I noticed the amount of difference recorded was varying with time - It goes from nearly the loudness of a normal file, to nearly silent, and back to the original loudness again. I think there is something in the setup causing this long-term variance. Or I just suck at Diffmaker. In any case, the plan was a failure.
Part 3: Stereo content

So I enabled SXFi

...

And disabled it

And enabled it again...

...

And disabled it again

I get that the goal is to reproduce speakers in a room. But if you're reproducing cheap speakers in a large hall, I'd rather you not.

A large part is related to that frequency response. (Keyword: Related, but not necessarily caused by. The FR itself may be the result of that other root cause.) The "cheap" sound. It is unfortunate that I don't have any SXFi-certified headphone / earphone. I only got, like, some obscure models from unknown brands like Alessandro MS-1 and Klipsch X11, I mean what has Klipsch even done before, like manufacturing loudspeakers since 1946?

But looking at the FR data above, the selected headphone / earphone only contributes to a small fraction of the FR, the bulk of it is still due to the SXFi effect. And testing different selected headphone / earphone confirms this - Selecting different headphones / earphones results in a noticeable and possibly important difference, but the effect is still small compared to the whole SXFi effect.

The sound also feels more distorted (as in the high THD kind of distorted) and details are lost. Definitely sounds more hollowy and reverb-y.

More importantly, this is what the SXFi effect does: Soundstage is compressed towards the center and moved forward. But not all of the sounds get this treatment. Some are spread around you including to the back.

The normal headphone sound sounds like sound hard-panned to the left and right, and in terms of front-back positioning it feels like right in the middle. The SXFi does take the sound out of this position and moves it more outside, some more forward and some more around. But still feels like it is coming from the headphones.

And I'm not sure if this can already be achieved by other competing effects, or if it can be done without resulting in such a big change to the sound. It does shift the sound forward which is relatively unique, most likely thanks to the included surround-sound technology. One thing I will say, and will say again in the 3D part of the review, is that the SXFi effect is very aggressive.

Here's the thing. I have earphone / headphone. I have speakers. My earphone / headphone have sounds that sound like my speakers, except they sound like they are coming from the sides of my head but otherwise they sound pretty much the same. And I believe most of you reading will agree with this statement.

So, why did the creators of SXFi feel like they need to change the sound *THIS* much? This is not "experience thatís indistinguishable from a dedicated set of surround speakers" or "high-end multi-speaker system in a professional studio", to be blunt I don't know wtf this is. Maybe an attempt at the feeling of magical immersion beyond the normal speakers sound? Then they should have written that instead of trying to simulate speakers, because it gives less things for people to pick at, and less things to fail.

Btw just for the lolz, if you apply SXFi effect on speakers, you get a soundstage that is even more compressed into the center and even further away from you forward. Logical, because the sound of speakers is already more center and forward, unless you are lucky enough to have a big studio with speakers far apart.

Also, you can recreate the headphone sound on speakers by putting your head right in between the speakers and pointing them at you. And even without pointing them at you (i.e. leaving the speakers pointing forward) the effect is already pretty close. (because speakers can be considered point sources, after ignoring that the frequency response certainly changes with direction, and baffle effects, and cabinet depth... and multi-driver interactions... but for the most part, speakers can be considered point sources)
Part 4: 3D

A good analogy would be: "3D displays"
(More accurately, stereo displays, but most people call them 3D displays)

You know how 3D displays have solid science behind, but when you watch them they still feel weird? Like you have to know how to look at the screen then the 3D effect will work? Be it Nintendo 3DS display, or those double pictures that you have to cross-eye and overlap the image or something.

It's the same here.

Left and right is easy to differentiate, but front and back is going to be hard. So we look out for audio cues that suggest the sound is coming from the back. Like maybe more "round", "dark", "reverby", "hollowy" or something. This has been done in games. And it is being done here.

In other words, you need to know how to listen to virtualized surround sound in order to hear what is being done here.

But SXFi's effect is very aggressive, that's fairly unique. It also creates an image in front, this is also fairly unique. It also seems to do better with multichannel content (Less things to fake and more things to work with I guess), but it still causes a significant quality drop in sound quality. Again likely due to the high aggressiveness.

In the best case, I would call this sound an acquired taste. In the normal case, I would disable it. Because the cost to SQ is too great and not worth whatever benefit it brings. It is like taking a pair of awesome front stereo speakers, and cutting it up into 7.1 crap ones. (And it doesn't even produce real 7.1 but I'm just saying.) So it is beneficial for like 3.5% of the time, but when the movie or game is doing nothing and is just having smooth music and a person talking in front, you want good sound for that but you can't get it. How about 7.1 good speakers instead? I don't know how hard this would be, but this would be the holy grail.
Very nice and details review.
Thanks for the effort.
__________________
If you're still poor at 35, you deserve it!
He knew what he signed up for.
delceer is online now   Reply With Quote
Old 07-11-2018, 01:08 PM   #8
Supremacy Member
 
jasonhanjk's Avatar
 
Join Date: Nov 2001
Posts: 6,100
Part 1: Analog performance

Outputting at ~1.90Vrms

Unloaded
20Hz THD=0.0016% THD+N=0.029%
1kHz THD=0.00021% THD+N=0.00063%
10kHz THD=0.00083% THD+N=0.0011%

33 ohm resistive load on both channels
20Hz THD=0.0019% THD+N=0.029%
1kHz THD=0.00090% THD+N=0.0012%
10kHz THD=0.0014% THD+N=0.0017%

Output impedance
Around ~0.4 ohm @ 1.9Vrms 33 ohm, limited by Arta reporting down to only 0.1dB precision

I'll just take the result as less than 1 ohm, I'm sure other reviewers with their own websites and stuff and their audio analyzers can get a more accurate number.

RMAA summary


The noise numbers look a bit weird, the spectrum shows some weirdness which does not happen with Arta. Comparing the Arta spectrums, it is around 5dB worse than my Xonar DX, so it should still be getting -105dB or better.

Of course, having the same noise level as a computer soundcard is not a good thing for an earphone amplifier, because computers are meant to be connected to speakers with volume controls, while earphones are sensitive and connected to the output of the, erm, earphone amplifier, which means noise levels need to be way lower. Somebody once said the more meaningful measurement of noise is in terms of volts. But what if this amp also has a digital volume control inside? That can change the amount of noise in terms of volts. Noise is not easy to quantify (hence why sometimes you can hear noise coming from a system that supposedly has -80dB or -100dB or better noise), and I'm feeling lazy. We shall see.

Results discussion
The numbers are unbelievable, but the graph checks out: A tiny bit of 3rd harmonic poking slightly above -120dB line.

One thing I noticed and remembered is that Arta and RMAA give slightly different numbers. And this is why I keep both measurements - Arta for the RMAA haters, and RMAA for comparison with other RMAA users.

I forgot I was supposed to use 16 ohm, not 33 ohm. But since I already did that might as well. I still have to collect max output power data anyway.

Back to the SXFi, unbelievable numbers. I didn't know my Xonar STX can measure this low, and because of this these numbers may actually be the limit of the measuring soundcard and not the SXFi.

The goal is to verify Creative's claim of THD+N=0.00036% IMD=0.0022% and crosstalk=-75dB and so far it seems to check out. I give an allowance of up to 6dB, if my numbers are worse than the claims (as is usually the case) it can be due to settings and equipment. Especially when THD+N of 0.0006% is the lowest that I have actually seen from Xonar STX self-loopback.

-75dB crosstalk without load is however pretty mediocre. Let's hope it is due to the cables/connectors/adapters, there is still chance for measurements. I have a feeling that this is the actual unloaded specs however. In which case it is really really bad, I mean because lots of cheap stuff can do at least this good with load.

I'm going to assume the SXFi should hit all the claimed specs. These are amazing numbers. The amp is amazing too, taking just a minor hit in performance from a 33 ohm load at slightly over 100mW in output power. LG has a smartphone that can hit the same number, but it doesn't have as much output power. This level of performance can win many desktop setups, and is packed into the tiniest package possible. Poor crosstalk requires further investigation, would be interesting to check if it is power supply related.
300 ohm headphone able get 12mW power.
__________________
Ask me anything about manicure or JB. :p
jasonhanjk is offline   Reply With Quote
Old 07-11-2018, 03:20 PM   #9
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
300.ohm.headphone.able.get.12mW.power.
Well yea but I wasn't testing at max output power or voltage

I tested at 1.9V because that is close to the redbook standard of 2Vrms and coincidentally/consequently the input voltage of Xonar STX at 0dBFS.

Based on numbers provided by Creative:

Load(ohm).Power(mW).Voltage(V).Current(I)
16.............425............2.607680..0.1629800
32.............370............3.440930..0.1075290
300............51.............3.911521..0.0130384
600............26.............3.949683..0.0065828
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"
wwenze is online now   Reply With Quote
Old 09-11-2018, 10:13 PM   #10
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
I created a new type of measurement thanks to this amp...

I call it: THD with white noise on the other channel

Because this thing has such disappropriational and abysmal crosstalk performance, I realized that this has no impact on sine tone testing since both channels are reproducing the same thing, but in actual usage the sound on the other channel is going to leak over and degrade the performance.

So for this test, I make the other channel produce white noise of the same RMS.

Result: It degrades the THD+N performance to pretty sad levels, mainly because of the extra noise.

33 ohm:


Note the much higher noise floor of the yellow graph as a result of white noise from the other channel leaking over.

The observant eyes would notice the green graph noise floor is worse than in the other measurements. This is because I have both soundcards on the some computer and so this is the common-mode noise. Still lower than the noise from crosstalk.

On the plus side, crosstalk seems to be lower when powered by the computer. RMAA reports around -59dB with a 16 ohm load, albeit not at 1.9V (slightly lower voltage was used since SXFi AMP cannot output 1.9V with a 16 ohm load).
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"
wwenze is online now   Reply With Quote
Old 10-11-2018, 09:44 AM   #11
Senior Member
 
Join Date: Dec 2001
Posts: 1,832
I created a new type of measurement thanks to this amp...

I call it: THD with white noise on the other channel

Because this thing has such disappropriational and abysmal crosstalk performance, I realized that this has no impact on sine tone testing since both channels are reproducing the same thing, but in actual usage the sound on the other channel is going to leak over and degrade the performance.

So for this test, I make the other channel produce white noise of the same RMS.

Result: It degrades the THD+N performance to pretty sad levels, mainly because of the extra noise.

33 ohm:


Note the much higher noise floor of the yellow graph as a result of white noise from the other channel leaking over.

The observant eyes would notice the green graph noise floor is worse than in the other measurements. This is because I have both soundcards on the some computer and so this is the common-mode noise. Still lower than the noise from crosstalk.

On the plus side, crosstalk seems to be lower when powered by the computer. RMAA reports around -59dB with a 16 ohm load, albeit not at 1.9V (slightly lower voltage was used since SXFi AMP cannot output 1.9V with a 16 ohm load).
Hmm....is it normal to have such amps to have such high crosstalk? Any previous measurement you did for similar amps to compare and contrast against the SXFi?

And probably a noob question: the spike at the 1K, is that because the test tone is 1K?

Last edited by lxXXxl; 10-11-2018 at 09:47 AM..
lxXXxl is offline   Reply With Quote
Old 10-11-2018, 09:57 AM   #12
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Hmm....is it normal to have such amps to have such high crosstalk? Any previous measurement you did for similar amps to compare and contrast against the SXFi?

And probably a noob question: the spike at the 1K, is that because the test tone is 1K?
I would say -60dB with a 33 ohm or 16 ohm load is "good", but with so many good products nowadays this "good" is now average or normal.

For my measurements:
http://wwenze.blogspot.com/2012/02/f...n-work-in.html



So SXFi doesn't do better than these.

Crosstalk SXFi + Win10 tablet:
Unloaded: -77.3dB
33 ohm: -50.3dB
16 ohm: -45.4dB

Crosstalk Honor 8 Pro @ -16.9dBFS of STX:
16 ohm: -44.2dB

Crosstalk SXFi + Honor 8 Pro @ -17.3dBFS of STX:
16 ohm: -42.9dB

The spike at 1k is because the test tone is 1k

Some random 16 ohm crosstalk numbers from the internet:
O2 amp: -65dB
SMSL M3: -61dB
Fiio E06: -57.3dB
Fiio A1: 63.7

TBH I don't know which amps on that website are as small as the SXFi, but losing to the cheaper Fiio is pretty...

But how about losing to Redmi Note 4: -58.9dB

And this is why I say SXFi's crosstalk is just average. Compared to other "free" products. And abysmal if you're paying $200 for it if the SXFi doesn't work out for you.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 10-11-2018 at 11:44 AM..
wwenze is online now   Reply With Quote
Old 10-11-2018, 11:20 AM   #13
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Also, there is a reason why I weigh noise and crosstalk much more heavily than distortion for my verdict.

Well, firstly we have to understand the diminishing returns. Beyond a certain level of performance, we can't hear it anyway. But this also means poorer than a certain level of performance, we can hear it and what we hear gets worse if the performance gets worse. So the starting few dB of performance is more important than the last few dB, and it is more important to have an all-rounded performance than excellent in one but poor in another.
How low humans can hear versus the main signal is always being debated. There are estimates of -40dB (or 1%) to -60dB (or 0.1%).

And more importantly, when you reduce the volume:

- Noise gets worse
- Distortion gets better
- Crosstalk stays around the same (can be better can be worse)
- THD+N gets worse

And distortion getting better is mostly due to the distortion becoming lower than the noise floor. I mean, say the distortion is -80dB (0.01%) but your signal is only 70dB above the noise floor. The distortion is smaller and buried by the noise. But THD+N gets worse because of the N.

And because we don't always listen at high voltages, we are going to be using lower voltages where noise is going to be the bottleneck and distortion is too low to be heard. This is especially true for IEM and pretty much anything with high sensitivity in terms of dB/V.

And so the below is the comparison of Honor 8 Pro 3.5mm and SXFi at my usual listening level, with a Klipsch x11 connected (Rated impedance 50 ohm):


Crosstalk:
3.5mm: -49.9dB
SXFi: -47.1dB

The 3.5mm does have higher distortion. However that peak at 3kHz is still more than 60dB away from the signal. On the other hand, the SXFi has worse THD+N numbers. Because the RMS of the noise is greater.

The same trend continues even at higher output volume. Below is the max volume of Honor 8 Pro 3.5mm, with the SXFi compared at the same volume. Noise is still the main reducer of performance.



Which means, unless you need the higher output voltage of SXFi,

The SXFi does not give better amp performance than a smartphone, at least with lower impedance headphones.

Except one thing I almost forgot: Output impedance of Honor 8 Pro is ~3.9 ohm. Most phones tend to not have low output impedance so that is one advantage the SXFi AMP retains.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"

Last edited by wwenze; 10-11-2018 at 11:45 AM..
wwenze is online now   Reply With Quote
Old 10-11-2018, 12:13 PM   #14
Senior Member
 
Join Date: Dec 2001
Posts: 1,832
Is the reduction of crosstalk a matter of additional circuitry? Or better quality components? Maybe that's a tradeoff Creative did to keep the amp smaller? Or cheaper to improve the profit margin, given the focus is on the SXFi mode.

For example, the Dragonfly's measured crosstalk is better than the SXFi's:
http://archimago.blogspot.com/2017/0...black.html?m=1

Can we compare device measurements like that? What variables do we need to be mindful of? Different tester conditions and equipments?

Last edited by lxXXxl; 10-11-2018 at 12:22 PM..
lxXXxl is offline   Reply With Quote
Old 10-11-2018, 02:32 PM   #15
High Supremacy Member
 
wwenze's Avatar
 
Join Date: Dec 2002
Posts: 26,177
Can we compare device measurements like that? What variables do we need to be mindful of? Different tester conditions and equipments?
Yes and no, easier to point things out as we go along. Measurements are only valid for that particular set up, but this also means if another person has the same set up, he is bound to get the same results. This is why you see people measure different setting conditions, some even doing multiple shmoos of graphs.

For example, the Dragonfly's measured crosstalk is better than the SXFi's:
http://archimago.blogspot.com/2017/0...black.html?m=1
I don't see Archimago saying he used 16 ohm or whatever ohm load with the DF Black anywhere. So the results are not comparable. However, I do have the unloaded measurements of SXFi. The Dragonfly's crosstalk is worse, since it measures at -71.5dB unloaded while SXFi measures at -75.6dB.

However both are simply mediocre compared to the rest in this chart.



And usually crosstalk just gets worse with load. If the Dragonfly is losing to SXFi unloaded, it is likely to also lose to it when loaded.

Is the reduction of crosstalk a matter of additional circuitry? Or better quality components? Maybe that's a tradeoff Creative did to keep the amp smaller? Or cheaper to improve the profit margin, given the focus is on the SXFi mode.
Crosstalk is usually due to weak power supply, particularly the ground. Better ground routing and more filtering tends to help. This results in an increase in size.

However, the SMSL iDEA is also small, yet has crosstalk at the same level as others.

https://archimago.blogspot.com/2017/...a-usb-dac.html

Plus the SMSL iDEA's noise is also better than the rest, while SXFi's noise is worse than the thing it is connected to. Again, weak power supply. Or rather, weak power supply noise rejection too.

This reminds me of the case of higher-performing IC vs lower-performing IC. The LM4562 vs NE5532 is an example. LM4562 gets higher performance, but if you just throw it into a circuit that uses NE5532, you end up with more noise. If you throw a fast digital circuit e.g. DAC without sufficient decoupling, you end up with more noise too.

And this is what I believe is happening with SXFi and AK4377. A high-performing DAC was chosen for the good distortion numbers, but higher noise as a result. The overall result is actually worse than if a lower-performing DAC were to be used, as explained in post #13. This doesn't change the fact that the numbers look amazing on marketing material, regardless of practical performance. And I feel this is what they are aiming for.

While part cost does matter, looking at how SMSL iDEA is doing, the problem is probably design. And/or profit margin.
__________________
"To learn who rules over you, simply find out who you are not allowed to criticize"
wwenze is online now   Reply With Quote
Reply
Important Forum Advisory Note
This forum is moderated by volunteer moderators who will react only to members' feedback on posts. Moderators are not employees or representatives of HWZ. Forum members and moderators are responsible for their own posts.

Please refer to our Terms of Service for more information.


Thread Tools

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are On