Is it possible to measure the difference between the referenced CD signal and the reconstructed MP3 from different bit-rates and report the difference as THD?
it would be a mistake - the whole point of psychoaoustic lossy compression is to use a model of human hearing and throw out the parts of the music signal that would be masked - these are usually the lower harmonics - so you would expect "High THD" from a properly working lossy CODEC with a complex, harmonically dense source like most music
for single or few widely spaced sine tones the compression will likely preserve the spectrum, most psychoustic model based compression will not introduce new distortion harmonics above the numerical precision of the decoder/possible widowing artifacts of the DFT
http://www.diyaudio.com/forums/anal...-3v-1khz-sine-wave-generator.html#post2786744
for single or few widely spaced sine tones the compression will likely preserve the spectrum, most psychoustic model based compression will not introduce new distortion harmonics above the numerical precision of the decoder/possible widowing artifacts of the DFT
http://www.diyaudio.com/forums/anal...-3v-1khz-sine-wave-generator.html#post2786744
Last edited:
Seems to me an altogether better way would be to subtract the mp3 from the original and report the difference signal. In other words, a null test. THD can only be measured on single tone signals as it relies in a very narrow notch filter.
again ignores what the lossy compression is trying to do - it is designed to leave out (= add 100% distortion to cancel to 0) lots of the signal - that you wouldn't perceive in the presence of the parts that it keeps
the compression is working if there is a big (visible) signal difference - that you can't hear - when compared to the original
the only way to judge the quality of psychoacoustic compression is to compare the playback - by ear - to the original
technical measurements make no sense unless you understand the Codec algorithms, psychoaoustic models and want to measure the implementation accuracy to the model
the compression is working if there is a big (visible) signal difference - that you can't hear - when compared to the original
the only way to judge the quality of psychoacoustic compression is to compare the playback - by ear - to the original
technical measurements make no sense unless you understand the Codec algorithms, psychoaoustic models and want to measure the implementation accuracy to the model
Last edited:
Yep - so how does what I said 'ignore what lossy compression is trying to do' please?
<edit> Delete the leading 'yep' as jcx has added some irrelevancies to his original post. Question still stands.
<edit> Delete the leading 'yep' as jcx has added some irrelevancies to his original post. Question still stands.
Last edited:
Hi,
THD+N is a brute force subtractive number that is easy to
measure but tells you next to nothing, unless its very low.
Its totally meaningless applied to non-linear systems like codecs,
as its based on a sliding single tone test, and its fairly obvious if
applied MP3 would do fine until it runs out of bits at the top end.
rgds, sreten.
THD+N is a brute force subtractive number that is easy to
measure but tells you next to nothing, unless its very low.
Its totally meaningless applied to non-linear systems like codecs,
as its based on a sliding single tone test, and its fairly obvious if
applied MP3 would do fine until it runs out of bits at the top end.
rgds, sreten.
Last edited:
I see measurement of THD can't be made without using test tones.
Not even if FFTS are employed. The distortion components in THD differ from codec information removal.
Not even if FFTS are employed. The distortion components in THD differ from codec information removal.
Last edited:
Hi,
THD in simple terms cannot be measured on its own.
Its always effectively THD+N in a bridged nulling circuit.
I don't see what FFT's have to do with it.
rgds, sreten.
THD in simple terms cannot be measured on its own.
Its always effectively THD+N in a bridged nulling circuit.
I don't see what FFT's have to do with it.
rgds, sreten.
Last edited:
If you use a plurality of tones (more than one) then it becomes an IMD measurement, no longer THD.
The FFT might well be useful to analyse the output - if it shows the original tones they can be recognized and set to zero, Its still IMD though and not THD.
FFT's would reduce the music into it's component tones. Then, each component tone and harmonic could be analyzed. eg. Take a 1Khz component (if any) from the reference and mp3 and all harmonics from both sources. Subtracting the harmonics of the reference from the mp3, would leave the the error. These could be summed.
Oh, I think I see. Is effectively what the nulling circuit is doing?
THD is only a useful number (and not hugely useful even then) when the output of the system is intended to be a copy of the input to the system. This is not the case with MP3 and other lossy encoders: they aim to produce a sound which is sufficiently similar to the original sound for the purposes of human ears/brains only.
You could put an MP3 system through a distortion analyser and get a THD number out but it won't contain any useful information. It will provide the illusion of knowledge, at best.
You could put an MP3 system through a distortion analyser and get a THD number out but it won't contain any useful information. It will provide the illusion of knowledge, at best.
Is that Bistromathics? (seriously the increase in noise floor or THD+N numbers provide another ring in the bathtub of hearing limits)
Pretty much, as in we know the answer, and now need to figure out what the actual question is.
Sure, but in the case of perceptual codecs such as mp3 somebody took out the bathtub and replaced it with a shower stall.
- Status
- Not open for further replies.