Analyzing Samsung UHQ Upscaler with measurements (on Galaxy Note 8 Exynos)

Which UHQ Upscaler mode do you use on your Galaxy S8/S8+/Note8?

  • Everything disabled

    Votes: 12 50.0%
  • Bit upscaling

    Votes: 9 37.5%
  • Bit and bandwith upscaling

    Votes: 3 12.5%

  • Total voters
    24
post-14076367
Thread Starter
Post #1 of 23

rafaelpernil

Head-Fier
Joined
May 28, 2013
Messages
55
Reaction score
52
Joined
May 28, 2013
Posts
55
Likes
52
Hello guys, since I've bought my Galaxy Note 8, I have been extensively trying each and every sound option available on settings. And I saw this one called Samsung UHQ Upscaler.

I looked up some info about it and it seems to upscale all audio to 32 bits, but I found two options underneath this upscaler: Bit upscaling and Bit and bandwidth upscaling

Trusting my ears I found out to be sounding the best when I only enabled Bit upscaling, but I wanted to see what this was doing, so I took measures with my Focusrite Scarlett 2i2 (2nd gen) connected to my iFi micro iUSB3.0 and RMAA 6.4.1. I also adjusted the input gain when needed to maintain the volume constant.




As you can see in this summary, there is an absurdly abrupt amount of Intermodulation Distortion and Total Harmonic Distortion when bandwidth upscaling is applied. But let's look at the graphs...



No variation on the frequency response, great!


Bit/Bandwidth mode has lower noise but also has some strange peaks arround 200Hz, 400Hz and 750Hz and Bit mode has lower noise without strange peaks.


Now we have a worse dynamic range on Bit/Bandwidth mode. And a slightly better dynamic range on Bit mode compared to having UHQ completely disabled. But it is within margin of error.


Here comes the disaster. Bit/Bandwidth mode is generating tons of harmonic distortion! Bit mode and disabled have the exact same distortion (result within margin of error).



And if that weren't enough, Intermodulation distortion on Bit/Bandwidth mode is crazy high! And again, same results on Bit mode and disabled, if anything, slightly better on bit mode.



Aaand here we have an interesting result. Bit/Bandwidth is giving the worst result by far doing weird things with sub bass frequencies. With UHQ Upsaling disabled, we get better crosstalk up to 5kHz, when Bit Upscaler keeps it lower than -90dB. Amazing!


I have repeated this test a few times and the results were consistent.


Analyzing the results, I see bit upscaling is really making some improvement. And we completely discard Bit/Bandwidth mode.

But now I want to check how does it perform on low volume. Because, as you may know, Android regulates the volume on a digital manner, reducing the overall available dynamic range (truncating PCM packets)



As you can see, we get lower noise, 6dB MORE OF DYNAMIC RANGE, lower THD, lower IMD and better crosstalk.



Exact same frequency response



UHQ Upscaler on bit mode shows a lower noise level specially on higher frequencies. Awesome!



An enormous difference in dynamic range, a whole bit is recovered with 6dBs more of DR.



Significantly lower Total Harmonic Distortion with UHQ Upscaler on Bit mode



Again, better Intermodulation distortion with UHQ enabled on Bit mode



And significantly better crosstalk overall.


Now, my conclusions about this are that indeed UHQ Upscaler on Bit Upscaling mode is the best option objectively talking. It seems to help quantize giving more resolution and lower noise. And I can lower the volume finally without loosing almost any quality.

Thanks for reading. What are your thoughts about it?
 
Last edited:
     Share This Post       
post-14076423
Post #2 of 23

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Messages
21,385
Reaction score
3,485
Location
Hollywood USA
Joined
Nov 16, 2004
Location
Hollywood USA
Posts
21,385
Likes
3,485
Website
www.facebook.com
Nothing there appears to be in the audible range for normal music listening. Have you tried doing a blind comparison?
 
Last edited:
     Share This Post       
post-14076449
Post #3 of 23

rafaelpernil

Head-Fier
Joined
May 28, 2013
Messages
55
Reaction score
52
Joined
May 28, 2013
Posts
55
Likes
52
Nothing there appears to be in the audible range for normal music listening. Have you tried doing a blind comparison?
My typical music listening scenario is more like the second one, at low volume. And there, I really notice the difference in noise floor and detail when changing between Bit upscaling and no effects. But I might try making a blind test to confirm it.

I wanted to give my grain of sand to this topic since I have not found any information about the actual effect of this function.
 
     Share This Post       
post-14076722
Post #4 of 23

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Messages
21,385
Reaction score
3,485
Location
Hollywood USA
Joined
Nov 16, 2004
Location
Hollywood USA
Posts
21,385
Likes
3,485
Website
www.facebook.com
You can hear a 6dB difference in a noise floor around -100dB at a low volume level? Are you sure something else isn't creating the noise?
 
Last edited:
     Share This Post       
post-14077044
Post #5 of 23

rafaelpernil

Head-Fier
Joined
May 28, 2013
Messages
55
Reaction score
52
Joined
May 28, 2013
Posts
55
Likes
52
You can hear a 6dB difference in a noise floor around -100dB at a low volume level? Are you sure something else isn't creating the noise?
Well, I guess I have to make a blind test and analyze the results. Thanks for the input!
 
     Share This Post       
post-14078412
Post #6 of 23

Zapp_Fan

Member of the Trade: Aiwa
Joined
Aug 29, 2017
Messages
363
Reaction score
180
Location
Chicago, IL
Joined
Aug 29, 2017
Location
Chicago, IL
Posts
363
Likes
180
Website
www.aiwa.us.com
So, I might be missing something, but changing a digital audio signal from 16 bits to 32 shouldn't affect the output if the original file was 16 bits. You have the same number but basically add more zeros after the decimal point (not literally, but that's more or less the result).

The fact that you got different results after upscaling (whether or not it's audible, I am not going to argue right now) to me indicates that maybe there is some problem at the firmware or OS level. In other words, maybe some other step of the phone's DSP is creating distortion / noise that varies with bit depth. Again, I'm assuming the upscaling itself has no direct benefit. THD changing with bit depth is not something I have seen before. Noise could theoretically improve if you switched the output mode of the DAC, so that sort of makes sense, but the rest has me scratching my head.

Someone correct me if I'm wrong, but I was under the impression that upscaling to 32 bit technically shouldn't make any difference at all, assuming the source signal is the same.
 
Last edited:
     Share This Post       
post-14078475
Post #7 of 23

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Messages
21,385
Reaction score
3,485
Location
Hollywood USA
Joined
Nov 16, 2004
Location
Hollywood USA
Posts
21,385
Likes
3,485
Website
www.facebook.com
Why there is a difference is the question you ask after you determine that there actually is a difference.
 
     Share This Post       
post-14078483
Post #8 of 23

Zapp_Fan

Member of the Trade: Aiwa
Joined
Aug 29, 2017
Messages
363
Reaction score
180
Location
Chicago, IL
Joined
Aug 29, 2017
Location
Chicago, IL
Posts
363
Likes
180
Website
www.aiwa.us.com
Why there is a difference is the question you ask after you determine that there actually is a difference.
True story. Looking at the graphs I initially assumed the measurements were indicative of a real performance difference. However, looking again, all of the changes seem to be roughly of the same dB difference. Maybe there is a funny scaling factor in there somewhere.

Anyway, if the differences are real, they are big enough to deserve some explanation, whether or not they are audible.
 
     Share This Post       
post-14078567
Post #9 of 23

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Messages
21,385
Reaction score
3,485
Location
Hollywood USA
Joined
Nov 16, 2004
Location
Hollywood USA
Posts
21,385
Likes
3,485
Website
www.facebook.com
Down that far, could it just be pushing the limits of what the testing equipment can measure?

I generally don't care about inaudible sound. I've got my hands full with improving audible sound.
 
     Share This Post       
post-14078584
Post #10 of 23

Zapp_Fan

Member of the Trade: Aiwa
Joined
Aug 29, 2017
Messages
363
Reaction score
180
Location
Chicago, IL
Joined
Aug 29, 2017
Location
Chicago, IL
Posts
363
Likes
180
Website
www.aiwa.us.com
I'll have to let raphaelpernil speak to the sensitivity of his gear, but the differences look so consistent that to me it suggests a common underlying cause.

I also don't care about inaudible sound. On the other hand, I kind of just enjoy this sort of discussion on an academic level. I know it makes no difference in real life, but it makes me a bit curious. This is the Sound Science forum after all, not the Practical Applications of Sound Engineering forum :wink:
 
     Share This Post       
post-14078681
Post #11 of 23

bigshot

Headphoneus Supremus
Joined
Nov 16, 2004
Messages
21,385
Reaction score
3,485
Location
Hollywood USA
Joined
Nov 16, 2004
Location
Hollywood USA
Posts
21,385
Likes
3,485
Website
www.facebook.com
I'm part of team Practical Application of Science. I'm not so keen on the navel gazing side of science.
 
     Share This Post       
post-14079246
Post #12 of 23

rafaelpernil

Head-Fier
Joined
May 28, 2013
Messages
55
Reaction score
52
Joined
May 28, 2013
Posts
55
Likes
52
So, I might be missing something, but changing a digital audio signal from 16 bits to 32 shouldn't affect the output if the original file was 16 bits. You have the same number but basically add more zeros after the decimal point (not literally, but that's more or less the result).

The fact that you got different results after upscaling (whether or not it's audible, I am not going to argue right now) to me indicates that maybe there is some problem at the firmware or OS level. In other words, maybe some other step of the phone's DSP is creating distortion / noise that varies with bit depth. Again, I'm assuming the upscaling itself has no direct benefit. THD changing with bit depth is not something I have seen before. Noise could theoretically improve if you switched the output mode of the DAC, so that sort of makes sense, but the rest has me scratching my head.

Someone correct me if I'm wrong, but I was under the impression that upscaling to 32 bit technically shouldn't make any difference at all, assuming the source signal is the same.
Good observation. I suspect the SRC (Sample Rate Converter) of Android is what's making the difference appear. Doesn't have any good fame

I will redo the tests diferenciating AudioTrack (default) and OpenSL ES audio APIs
 
     Share This Post       
post-14129810
Post #13 of 23

Duncan

Headphoneus Supremus
Moderator
Joined
Jun 26, 2001
Messages
13,147
Reaction score
1,175
Joined
Jun 26, 2001
Posts
13,147
Likes
1,175
Is this on Exynos or SnapDragon based phone? Exynos has a separate DAC, SnapDragon uses inbuilt...

Thanks.
 
     Share This Post       
post-14143696
Post #14 of 23

swesko

500+ Head-Fier
Joined
Mar 27, 2016
Messages
626
Reaction score
103
Joined
Mar 27, 2016
Posts
626
Likes
103
Wish I had your tools to do the same thing for my exynos s9+. I have a DAC but I might sell it as my s9 sounds better than my s8 plus is he a a mojo I can always use.
 
     Share This Post       

Users Who Are Viewing This Thread (Users: 0, Guests: 2)

Top