MP3 player distortion measurement [shocking!]

[Limhes]

Few weeks ago I bought my first MP3 player, just to have some music on the road (bike, train). It sounds awful. Really bloody cracky and everything you could ever imagine. So, I thought my ears were too sensitive or something, and started measuring.

So, just made a quick frequency response of my MP3 player versus a cable between sound card in- and outputs:

Soundcard:


MP3 player:


I didn't know it could be this awful and I am scared of even approaching a reasonable guess for its THD...

So, my question:
What do these guys put in a device like this to make it this horrible???

[jcx]

Originally posted by Limhes
...
So, my question:
What do these guys put in a device like this to make it this horrible???

Psychoacoustic lossey compression is a nonlinear process - the “distortion” results are readily understood as expected from the details of the compression algorithms

By exploiting a variety of masking phenomena, nonlinear loudness perception and "spending" bits only to encode what features make a difference in what people hear better audio compression schemes can "toss out" >75% of the data in a 16/44.1 CD with a vanishingly small number of people able pick to out the difference in blind testing

see the tech papers section of
http://www.mp3-tech.org/

[Limhes]

I understand, but aren't you talking about MP3 (or lossy) compression in general? I know how these work, generally, involving acoustic masking.

I used an MP3 file (the same!) in both cases. The distortion figures are from my portable MP3 player only! Now, this player just takes the files I give it and plays them, but why would they ever distort these files even more? You think it's a loudness algorithm inside the device?

[jcx]

I suspect some automated tests could look bad based on a mls/noise or multisine test where the compression algorithm is busy exploiting psychoacoustics to remove "unhearable" features from the test signal

A slow single sine sweep passed through the mp3 encode/decode chain should give a flat frequency response, likewise single tone test sine should show no added harmonics - within the arithmetic resolution limits

your second plot shows classic square wave harmonic structure - as a guess you may be severely mismatching levels and clipping badly

I'd expect few portable mp3 players would saturate a typical soundcard line in - check to see if you're not accidentally configured for mic in or just reduce the test tone level to see if you get an abrupt improvement as you quit clipping

[Notagenius]

i'm having a hell of a time with compressed Morse Code. it has all sorts of ringing.

[I_Forgot]

Originally posted by Notagenius
i'm having a hell of a time with compressed Morse Code. it has all sorts of ringing.

Is the ringing due to a narrow band CW filter or due to the mp3 encoding process?

I_F

[hehe31]

How do you make this test? (which soft to generate the test wave file and to see the result)
i'll try with my Sony "Walkman" phone

[Notagenius]

sorry for threadjacking.

I_F,

http://www.arrl.org/w1aw/morse.html

the mp3s were played back on xmms on linux. i'm going to try some other players tonite. tnx.

[I_Forgot]

Originally posted by Notagenius
sorry for threadjacking.

I_F,

http://www.arrl.org/w1aw/morse.html

the mp3s were played back on xmms on linux. i'm going to try some other players tonite. tnx.

It sounds crisp here, no ringing at all.

I_F

[janneman]

I think what you see in the 'distorted' freq response is the MP3 encode/decode process at work. Only spend bits where it counts, and leave out what would be masked anyway. Naturally this looks non-linear. If you use another sound file, it should look completely different.

Jan Didden