Technical Description Digital Filters / Oversampling - T+A MP 3100 HV User Manual

Technical description
Digital filters / Oversampling
Oversampling The audio data on for example CDs is stored at a sampling rate of 44.1 - i. e.
for each second of music 44.100 sampled values are available for each
channel. In the MP 3100 HV the audio data read from the CD is „multiplied" to a
higher sampling rate (352,8 kHz) before it is converted back into analogueue
music signals. This process delivers a very much better, more finely graduated
signal to the converter, which can then be converted with correspondingly
higher precision. The raised sampling rate is a calculating process for which
there are many different mathematical methods. In almost all digital audio
devices which exploit the advantages of increased digital sampling rate a
process known as a FIR filter is employed for this purpose. At  we have
been carrying out research for more than ten years, aimed at improving the
oversampling process, because the standard FIR method has one drawback to
set against its indisputable advantages: it adds small pre- and post-echoes to
the music signals. At  we have developed mathematical processes (known
as Bezier polynomial interpolators) which do not share this disadvantage. For
this reason they should sound better and more natural than the usual standard
process. Since the calculating procedure employed by us is considerably more
complex than the standard method, the MP 3100 HV features a high-
performance digital signal processor (DSP) which carries out the over-sampling
process with immense precision (56 bit) using special algorithms developed by
.
The freely programmable DSP which we use is capable of carrying out the
oversampling process using any method of calculation. For this reason we
have implemented a slightly modified Bezier process (filters 3) in the
MP 3100 HV in addition to the pure Bezier process (filter 4), together with two
variants of the standard process (filter 1 and filter 2). For more information on
the different processes please refer to the next section. You can switch
between the various algorithms, then decide for yourself which of the filters
gives the results you prefer.
FIR long (Standard FIR Filter)
The long FIR filter is the standard oversampling process in digital technology,
offering extremely linear frequency response, very high damping, linear phase
characteristics and constant group delays. The disadvantage is the pre- and
post-echoes which are added to the signal. These „time range errors" tend to
affect the music signal's dynamics, precision and naturalness, and reduce
spatial orientation.
Frequency response and transient characteristics of the long FIR filter
FIR short (Impulse optimised filter)
Shortening the filter (lower coefficient) reduces the time range errors, albeit
combined with a slight loss of linearity in the frequency range and damping
performance.
Frequency response and transient characteristics of the short FIR filter
59