reported
by: Mr. Rafael Oei
Secretary (Term
2000/2001)
On Wednesday 28 February 2001,
6 members and 12 guests of the AES Singapore Section gathered
in Lecture Theatre 63 at Ngee Ann Polytechnic.
Eric Chan, Senior
IC Design Engineer of Creative Technology Singapore promised to share his
research on multi-rate signal processing utilised in Creative Technology’s
sound cards.
|
Mr. Eric Chan
presenting his talk on "A Primer on Multirate Signal Processing in
Audio" at Ngee Ann Polytechnic - photograph
by Mr. Robert Soo. |
Shannon’s
Sampling Theorem provided the launch pad for the night’s talk. Using
frequency wave captures, Mr Chan laid the groundwork skimming
through the problems of sampling and aliasing under various conditions.
He then contrasted filtered samplings within time and frequency domains,
always careful to avoid the math so that his presentation remained on a
common platform for the audience.
After simplifying Shannon’s
Sampling Theorem for the lay audience, Mr Chan introduced the
first schematic based on Sigma Delta ADC/DAC. From |
graphic representations of an
analogue signal from this configuration, Mr Chan explained the
problems Resistor Capacitor Networks presented. These were not for
integration into “ICs”, given the analogue circuit design, the difficulty in
resistor matching during integration, a limited power supply of 3.3V and
2.5V and performance degradation. Mr Chan then went on to describe
Sigma Delta Converters, explaining why it was not ideal for Creative
Technology’s purposes as well.
|
Finally, Mr Chan came to the model that Creative Technology had
adapted for its PC cards; the multi-rate signal process that allows for
arbitrary sample rates. This addressed the problem of processing audio
signals of different sample rates to enable audio integrity on playback
from a limited system such as a PC. Reconciling clocking errors and
accumulated mismatches from input sources had to be addressed despite
circuit integration constraints. Mr Chan added that due to ratio
returns in traditional means of sample rate conversion, I/D ratios were
not practical. Non-integer ratios makes for |
Section
Chairman, Mr. Robert Soo (left) presenting the AES speaker's plaque to
Mr. Eric Chan after the talk - photograph by
Mr. Michael Teh. |
clumsy multi-stage system and
audio applications require a system with a minimum number of multiply
accumulate cycles. So sample rate conversion had to be based on an arbitrary
factor.
|
AES members and
guests with the speaker, Mr. Eric Chan (front row, 3rd from left) gather
for a group photo shoot after the talk at Ngee Ann Polytechnic -
photograph by Robert Soo using self-timer. |
Using more wave captures, Mr
Chan went on to explain his findings in detail, using comparison charts
to illustrate the differences in passband responses based on Linear
Interpolation, Sinc Function Interpolation, Remez Optimised Interpolation
and the “new” approach based on multiple constraints. Why does it work?
Although, Mr Chan admits, the model is not based on psycho-acoustic
analysis, the strength is in the result. Normal human ears can’t seem to
detect the shortfalls from the resultant decoded audio signals from this
configuration.
The audience then had a preview of
what consumers can expect from Creative Technology in the years to come.
That, however, is a different story. |
