Vorbis dev - Mar 2005 - Query on 24 bits fixed point implementation

Hi,

We are trying to implement 24bit fixed point arithmetic of Tremor code.
To convert the 32 bit code we are truncating the data to 24 bits.

Based on the above changes we compared the compliance of the 24 bit
fixed point Tremor code with 32 bit fixed tremor code. We are testing
our compliance with respect to the floating point code.

We observed that SNR of 24 bit code is less than that of the 32 bit
code. Which is expected due to the loss of accuracy because of the
truncation of 8 bits.

To give a brief analysis:
SNR with respect to floating point code
(for)					(SNR)
The 32 bit tremor code		14  (for most of the songs tested)
The 24 bit tremor code		10  (for most of the songs tested)

We are targeting an SNR of 12 for the 24 bit Tremor code.

It will be helpful if someone can suggest some way of getting better
accuracy in case of 24 bit arithmetic.

Thanks in advance.
Regards,
Valmick




Confidentiality Notice

The information contained in this electronic message and any attachments to this
message are intended
for the exclusive use of the addressee(s) and may contain confidential or
privileged information. If
you are not the intended recipient, please notify the sender at Wipro or
Mailadmin@wipro.com immediately
and destroy all copies of this message and any attachments.