Hi Everyone
> It also reduces the input slew rate. This doesn't matter if the
> system is perfectly linear, but do we know that it is?
>
> Regards,
> Allan.
As long as this thread is still active and talking about new audio
standards, this might peak some interest.
Interestingly it was thinking about the importance of slew rate and how
this relates to dynamics since that is what we perceive best that led me
on to an audio compression patent a few years ago. Basically my
thinking was that for low frequency high amplitude bass, not much
precision was needed to express slew rate. And, at high frequencies,
especially those close to the Nyquist rate, again not much sample to
sample precision is needed when you only have 2 samples per sine wave
(eg why reconstruction filters are important)!
What I found was that by taking the first derivitive of an audio/video
signal and expressing it in floating point (or log form) that keeping
full precision was not required and that only a simple integration was
required at the back end. You can kind of think of this as ADPCM for
free (if you are using a floating point processor), but with the base
constrained to 2.
But the even better part to this argument was that by being a linear
process and that differentiation can be considered to be a simple 2-tap
HF pre-emphasis filter, the data can be processed in-situ.
One of the VC33-DSK demos is in fact a 10 stage graphic equalizer built
from 10 cascaded IIR filters. Even with only 5 exponent, 1 sign and No
mantissa bits it is mighty difficult to discern artifiacts from a CD
input. But then again, for a purist, why not toss in the remaining 24 bits!
The bottom line here is that this somewhat reduces truncation noise
sensitivity in the ALU... But you will need to have a (TI) floating
point ALU.
BTW: Differential compression does NOT help when it comes to the
precision of a coeficient. That is, coefficient precision errors still
cause the poles and zeros to move slightly. But then again what does
this do? It tends to move the center frequencies up/down a little bit.
Interestingly though it does open up one other benefit. If you do want
to use extended precision coefficients, the underlying math operations
are somewhat simplified.
Best regards,
Keith Larson
+--------------------------------------------------------------+
| Keith Larson |
| Member Group Technical Staff |
| Texas Instruments Incorporated |
| |
| 281-274-3288 |
| k-larson2@ti.com |
|--------------------------------------------------------------+
| TMS320C3x/C4x/VC33 Applications |
| |
| TMS320VC33 |
| The lowest cost and lowest power 500 �w/Mflop |
| floating point DSP on the planet! |
| |
| Web: focus.ti.com/docs/toolsw/folders/print/tmdsdsk33.html |
| Code: www-s.ti.com/sc/psheets/sprc147/sprc147.zip |
+--------------------------------------------------------------+