12
6. HiBit-16 (tm) -putting 24 bits onto 16 bit digital tape.
A/D CONVERSION
STAGE
ANALOG INPUT
DIGITAL OUTPUT
HIGH FREQUENCY
RANDOM BIAS
Fig. 6.1- HiBiT-16 process
Fig. 6.2- Very low noise sinewave
input at -90dBFs
Fig. 6.4- Same signal after
16 bit truncation
Fig. 6.3- Same signal after
HiBit-16 process.
+1 LSB
-2 LSB
2
10
+2 LSB
16 BIT LSB SCALE, 100Hz
-1 LSB
0 V
msec
+1 LSB
-2 LSB
2
10
+2 LSB
16 BIT LSB SCALE, 100Hz
-1 LSB
0 V
msec
+1 LSB
-2 LSB
2
10
+2 LSB
16 BIT LSB SCALE, 100Hz
-1 LSB
0 V
msec
Although the DAT and CD are "16 bit" it is commonly known that 16 bit converters are really 15 bit or less
and you need a higher resolution converter to approach theoretical limitations of a 16 bit medium. But still
higher resolution is needed for more sound information, especially for fine details, reverbs and ambience
sounds. The 24 bit ADC provides the resolution, but the tape is still 16 bit. You can preserve this high
resolution by using the HiBit-16 (tm) process which by adding high frequency bias modulates the 16th bit in
order to achieve close to 24 bit resolution. It can be achieved thanks to almost ideal differential linearity of
the Workstation24 and its very high dynamic range. The program recorded using this method has more detail
than if it was "truncated" to 16 bit. The noise floor is now kept flat on the 16 bit level (-96 dB) while the detail
is heard through the noise. Truncation on the other hand would produce low level distortion and result in the
loss of low level ambience sounds. In the real world the noise floor of the program is nowhere near -96 dB,
usually around -70 dB for an analog mix and around -90 dB if going directly from a good mic preamp. In our
opinion there is a little benefit from using various "bit mapping" systems which focus on lowering the noise
floor to -110 dB at the expense of coloring the sound and boosting (plus modulating) the high frequency
noise. HiBit-16 is a linear process which does not affect the sound and does not modulate the noise floor
leaving it flat at -96 dB level. When using the process please keep in mind that its effect will be less audible
on a program with high noise floor because of the noise becoming a natural "dither" for the converter. In any
case however, the low level resolution will be preserved. Here are some graphs illustrating the idea:
