User Manual

ALF-DSP 88-U
-11-
2.4 Floating Point DSP
The ALF-DSP 88-U adopts analogue devices’ ADI SHARC audio processors, enabling 32-bit and 40-bit floating-
point processing, which can be compared to 40-bit float ing-point processing of other devices. Floating-point
processing provides prominent advantages for the users in terms of sound quality and usability.
Fixed-Point Processing Limitations
Fixed-point processing has its own disadvantages. If there is a significant change in gain, data loss or a more
severe error may occur, including clipping or distortion. For example, the processing of 24-bit fixed point-based
audio signal, in some cases, if you attenuate the signal to 42dB, the new signal only includes 17-bit information.
Due to gain attenuation, 7 bits of inform ation w ill be lost forever. Worse than that is the clipp ing distortion. Fo r
a signal nearly close to 0dBFS, the s ignal will be clipped at 0dBFS, and audio distortion will occur. Even if the
signal level is adjusted to below 0dBFS through post-regulation, the clipping has occurred, and the distortion still
exists. Fixed point processing can create some headroom above 0dBFS, but by doing so, some bits have to be
abandoned. For example, if a 12dB (2 bits) headroom is created, a 24-bit system will only have 22 bits.
Floating-Point Processing
On the contrary, by taking advantage of floating-point processing, no matter what the s ignal level is, all available
bits are uniformly distributed to the signals. Basically, the floating points use some bits as indexes to set up
general signal level and distribute the remaining bits to signals with independently stored level. As a result, no
matter wh at kind of level (from -200dB and 200dB below to 0dBFS above, the stored signal's accuracy is
optimized without clipping distortion. SHARC provides 32-bit and 40-bit accurate processing; through 32-b it
processing, 25 bits are distributed to storage signals no matter what its signal level is. This means that, based on
at least 1-bit low level s ignal, its accuracy is always sign ificantly superior to 24-bit fixed point processing.
Through expanded 40-bit accurate processing, 33-bit storage signals can be achieved.
Practical Significance
What's the practical sign ificance of floating-point processing for the users? The gain stages between multiple
modules can be ignored. If the signal level of a module is reduced by 50dB and is then restored to its original
value through another processing, data loss will not occur. In the fixed-point system, the users must check other
signal levels before sending it t o the A/D converter because all digital-to-analogue converters adopt fixed points.
In the DSP system, if you notice that your signal has been clipped before it is outputted and transmitted to the
digital-to-analogue converter, you may close it off immediately at the output section to correct the situation. If
using the fixed-point system, you would have to search each processing module to find the clipping source.