Bob and Jim raise an interesting real-world aspect of A-D conversion.
Perhaps a brief overview might be of interest to those encountering
these concepts for the first time.
Theoretically, an N-bit A/D converter can provide 2^N discrete
quantization levels or steps or numbers corresponding to some
specified analog input signal amplitude range. Since each quantization
level represents a change in magnitude of 2, we equate each bit with 6
dB of available dynamic range for the converter.
Example: a 4-bit ADC can output the binary numbers 0000 through 1111,
or 0 through 15 decimal. These 16 levels would all be usable except
for the effects of indecision in the conversion process which results
in the least significant bit (LSB) sometimes being correct and
sometimes being incorrect. On average, the correct digital value
"falls in the crack between levels" and produces quantization noise in
the ADC output. That is, the numbers differ from sample to sample
from the "exact values" by being forced to take on only the available
values from the converter. This can be thought of as the true sampled
signal accompanied by essentially uniformly distributed noise across
the spectrum of the "real" signal.
Thus, we regard the LSB as being "lost" to the effects of quantization
noise even under the best of conditions. So, our N-bit ADC functions
at best as an (N-1)-bit converter, and we have lost 6 dB of its
theoretical dynamic range. The working dynamic range thus is reduced
from 4 bits at 6 dB per bit or 24 dB to 3 bits or 18 dB. Effectively,
we disregard the LSB and use only the remaining higher-order bits.
Realistically, more than just the LSB is lost to quantization noise in
the real world of A/D conversion. Common techniques such as sample
clock dithering to control the effects of quantization noise can "use
up" several low-order bits and therefore further reduce the effective
number of quantization levels based on the "6 dB per bit" viewpoint.
Apart from A/D issues, there are a number of other factors at work in
a digitally controlled receiver which influence overall dynamic range.
Chief among these is the analog response function of the circuitry
preceding the A/D converter. Non-linearities in circuit operation can
produce aggregate ADC input signal components of larger amplitude than
the simple combination of signal amplitudes that would result from
purely linear circuit operation ahead of the converter. Thus, more ADC
bits are required to represent the input signal than otherwise would
be needed with purely linear operation.
Effectively, then, about 100 dB of dynamic range "could" be associated
with about 16 or 17 active bits in the A/D conversion process, but
this may or may not accurately reflect the actual number of "working"
bit levels in the conversion process itself. Realistically, some
equivalent "bits" are lost to front-end non-linearities and some to
dithering and related ADC operational schemes.
Conclusion: it can be confusing and lead to misunderstanding to think
of the dynamic range of a digitally controlled receiver as being
determined solely and simplistically by the number of bit levels
provided by the ADC. Typically, a 24-bit ADC, theoretically capable of
24*6 = 144 dB overall numerical signal range, may produce a usable
dynamic range of only 100 dB as previously reported. The element of
design superiority plays into achieving as much of that theoretical
144 dB as possible in a production system; i.e., making the most of
the available 24 bits.
Doug Smith's very excellent chapter on DSP in the ARRL Handbook is
required reading for anyone interested in this subject area. Seldom
have so many complex topics been compressed into such small space and
explained so clearly. An outstanding contribution . . .
73/72, George
Amateur Radio W5YR - the Yellow Rose of Texas
In the 57th year and it just keeps getting better!
Fairview, TX 30 mi NE of Dallas in Collin county EM13qe
K2 #489 IC-765 #2349 IC-756 PRO #2121 IC-756 PRO2 #3235
----- Original Message -----
From: "Robert & Linda McGraw K4TAX" <RMcGraw@blomand.net>
To: "Jim Reid" <jimr.reid@verizon.net>; <tentec@contesting.com>
Sent: Friday, January 03, 2003 7:09 AM
Subject: Re: [TenTec] Orion vs. Jupiter Audio vs. DSP
> Jim brings up a very good point that most folks seem to overlook.
>
> That is "............ design seems to "only" realize some 18 or
maybe 19
> bits of "effective" A/D conversion as just over 100 dB dynamic range
is
> suggested by Doug as what is being realized in tests. Even if 32
bit A/D
> chips were available, it does not mean that that much resolution of
the
> analog conversion would occur. "
>
> State-of-the-art 24-bit IF-DSP converters, such as those used in the
Ten-Tec
> Orion, produce about 100 dB of dynamic range.
>
> 73
> Bob, K4TAX
>
|