oversampling causes the low-resolution quantizer to appear to have a much higher resolution.
This apparent higher resolution can be quantified by the ENOB and is found from
in order for the SNR to increase by approximately 6 dB. Each subsequent increase of 6
21
As seen from (1) and (2), to achieve an ENOB of 12 bits using a 1-bit quantizer,
a sampling rate
over 4 million times faster than 2f
max
is required. This obviously is not practical and shows that
ΣΔ converters must use other techniques in addition to oversampling.
The other key component in ΣΔ converters is the integrator that is placed before the 1-bit
quantizer. This integrator functions as a low-pass filter for the desired signals occurring at or
below f
max
and as a high-pass filter for the quantization noise in the ADC. This shapes the
quantization noise (which is normally flat across the band from 0 to f
s
/2) so that very little of this
noise occurs in the desired signal’s band (0 to f
max
). Most of the quantization noise is shifted to
frequencies above f
max
. This process is called noise shaping and is shown in Figure 9. The results
of this noise shaping are that the desired apparent resolution (ENOB) can be achieved with much
less oversampling than is predicted by (1) and (2).
Figure 9. Noise shaping in ΣΔ ADCs.
The effects of the integrator on the quantization noise of the ΣΔ converter can be seen
mathematically by considering a linearized model of the ΣΔ modulator portion of the converter.
The block diagram of this model is shown in Figure 10. The quantizer is modeled as a unity gain
amplifier with quantization noise added. Looking at this model in the frequency domain, the
output of the ΣΔ modulator Y(s) is given as
= [ ] (
1
)
where
X(s) is the input signal,
H(s) = 1/
s is the transfer function of the integrator, and
Q is the
quantization noise. This expression can be rewritten as
=
1
1
P
o
w
er
0 f
max
f
s
f
2
f
s
» 2f
max
Quantization Noise without
Noise Shaping
Quantization Noise with
Noise Shaping
P
o
w
er
0 f
max
f
s
f
2
f
s
» 2f
max
22
Figure 10. Linearized model of the ΣΔ modulator
This shows that at low frequencies (s « 1) the output is primarily a function of the input signal
Chia sẻ với bạn bè của bạn: