Frank - a technical question....

Frank McCoy

2005-02-09 20:32:30 UTC

Post by ÃldmanÂ©
Greetings Frank, how are you and yours?

Gee ... a relly *technical* question. ;-}
Doing fine I guess. Need some work though.

Post by ÃldmanÂ©
I was wondering if you knew a bit about mp3 conversion algorithms used. It
seems to me that it would have to be a Fourier transform of the sampled waveform
to encode then reverse the process to decode for 'playback'. I'm not having much
luck finding the more technical details of the process. When I do a search I get
thousands++ of hits for conversion software ect. and little in the way of technical
stuff.

I'm not sure. Methinks the algorithm uses wavelets, not Fourier
Transforms; but I could easily be wrong.

Post by ÃldmanÂ©
To save you a bit of time - I understand electronics, ie AD 'conversion' via
sampling the input signal, the 'ins and outs' of sample rates, 8-12-16 bit
AtoD etc. Also I have a fair bit of math (third year eng. level) and understand
Fourier transform functions [-- Fourier (fo rya; E fore er)
1 Francois Marie Charles (fran swa ma re sarl)
1772-1837; Fr. socialist & reformer
2 Baron Jean Baptiste Joseph (zan ba test zo zef)
1768-1830; Fr. mathematician & physicist
(C)1995 Zane Publishing, Inc. (C)1994, 1991, 1988 Simon & Schuster, Inc.]
I am particularily interested in "lossless" encoding as there is no way
that I am aware of to do that kind of encoding without losing something. The
only thing that approximates the analog signal is a wave file (in my mind).

It's true. All waveform encoding is lossy. That includes all movie
formats (MP*, REAL, Divx, etc.). MP3 is just a subset of MPG for the
audio portion. No, I don't know the algorithms.

Even wave files are lossy, usually. To record lossless wave files
often takes much more bandwidth than is available.

Post by ÃldmanÂ©
We 'sample' the analog signal and use an AD converter to develop a binary file
called a 'wave' file with sample rate being important as the higher the sample
rate (frequency) the better the binary represents the original analog signal.
We then use an alogrithm to 'sample' the wave file and develop a 'compressed'
file for easier transmission and storage, reversing the process when the
'compressed' file is 'played back' where another algorithm 'rebiulds' an
APPROXIMATION of the original wave file which results in loss of frequency
response - the lower the 'bit rate' the higher the 'amount' lost.

Not necessarily so. It depends more on the efficiency of the
algorithm than the bit-rate of the compressed file. The original
sampling-rate just sets the limit for which any compression hopes to
sound like on playback. And, it isn't usually frequency-response that
is lost, except in the original sampling; which is why you want
high-bit-rate samples.

Post by ÃldmanÂ©
All of this results in a 'loss' of frequency response, and if I remember
correctly our frequency range of hearing is 100 to 20 kHz. Above and below
our range of hearing we 'feel' the frequency rather that hear it.

Below, we feel it in our chests. Above, it depends on the person.
Some are completely deaf above some frequencies; others sense it's
there, but usually only at fairly high levels. No longer do we get
much of a sense of pitch though.

Post by ÃldmanÂ©
Also - most modern electronics (for the last 50+ years) is designed to provide
a 'bandwith' well above and below human hearing frequencies. This most particularily
applies to 'storage' media - ie: vinyl records, tape Cds and whatever.

Again, not really. *Amplifiers* generally are designed for extended
range, and with good reason; as such amplifiers (usually) have less
distortion as well.

However, *recordings* are generally very steeply and sharply cut off
below 16hz, and in most recordings below 20hz. They are also sharply
cut off above 20khz, and in most recordings above 16khz.

Again, good reasons for each.
Cutting off the unheard lows gives more total dynamic range to the
rest of the recording. To get the same sound-level with a bass-note
requires far more extension than the same mid-range note. This would
have to be graphed to be appreciated. With the high-notes, it has to
do with sampling rates and equipment. Mostly, with today's digitized
music, if you approach even 1/2 the sampling-rate, you can introduce
aliasing distortion; which is notes that just are NOT in the original
sound-source.

For lossless compression, you're pretty much limited to lossless
algorithms like Limpel Zev; the commonest used (and one of the most
efficient) being ZIP format by Phil Katz.

All other methods of storing sounds efficiently use deliberately lossy
algorithms. MP3 being one of the best, both for compression-ratio and
quality of finished sound. Generally though, you can figure the
better the compression ratio and sound quality, the more computation
needed by the processor doing the job.

That's why some video board support "on board MPG hardware
decompression"; to offload a lot of the work from the main computer
processor. On some *older* boards though, the "hardware* is slower
than the latest software running on newer computers with upgraded
processors.

Post by ÃldmanÂ©
PS - I have been 'out of school' for nearly 40 years so am not 'up to date'
on some of the stuff used in the digital world.

You might try this site as a start:
http://www.faqs.org/faqs/compression-faq/part1/section-20.html

--
_____
/ ' / ™
,-/-, __ __. ____ /_
(_/ / (_(_/|_/ / <_/ <_