For a normal A/D sampling at a rate approaching the Nyquist bandwidth (Fsample
/ 2) , Signal to Quantisation noise ratio is approximately 6.N - 1.7 - (peak to
mean ratio). Where N is the number of bits. So for the telephone network with
8kHz sampling, if we need 46dB S/N ratio for voice with a typical 10dB
peak-to-mean, then somewhere around 9 - 10 bits would necessary. So, to achieve
this performance on the telphone network using 8 bits / sample, non-linear
sampling (mu-law or A-law coding) is used which has the effect of compressing
then expanding the audio dynamic range to effect the increase in S/N ratio for
practical signals.
NOW, f you go to a lower number of bits but sample at a much higher rate, then
other tricks become possible. For example, delta modulation used many years
ago for a simple digital transmision over radio used 1 bit sampling,
transmitting a '1' if the amplitude has increased from the previous sample, and
'0' if it decreased. For speech a sampling rate of 30 to 50kHz gave acceptable
fidelity.
In modern Codecs this technique is carried further, but made transparent to the
user who sees only a 'classic' A/D converter operating at a lower rate. For
example, the codecs used in most soundcards actually sample the audio at many
MHz at a low resolution (1 or 2 bits often) in a simple but accurate A/D. Then
sucessive samples are combined digitally and processed to give an effective
sampling rate at the familiar values of 11025 - 44100Hz. The disadvantage of
this approach is that the A/D converter can never truely digitise a DC signal.
It can get arbitrarily close, but never down to 'true' DC. This is one
reason why a good Software Defined Radio using two channel I/Q processing
cannot be made out of a soundcard / PC combination. There will always be a
hole right in the middle of the spectrum corresponding to the DC response.
There is also the issue of capacitor coupling that reduces DC response on
practical hardware, but that's a matter for manufacturers.
The same techniques can be used at higher frequencies, but to achieve MHz
effective sampling rates needs GHz sampling speeds in the first place - which
is pushing sample-and-hold technology to its limits. The arguments are
exactly the same as described in earlier posts on this topic. When a wideband
signal is filterd and decimated (the sampling rate reduced) the S/N ratio of a
narrowband signal can be increased way beyond what would have been possible at
the original sampling rate.
Modern approaches can use combinations eg a high specification 4 or 8 bit
converter coupled with down sampling to give 16 bits or more. But what goes
on inside modern A/D chips is probably proprietory information and we see only
the outside, effective, conversion performance - and go 'WOW' when we see the
specs!
Andy G4JNT
On Thursday, February 24, 2005 11:24 AM, Claudio Pozzi [SMTP:[email protected]]
wrote:
On Thursday 24 February 2005 02:18, Alexander S. Yurkov wrote:
> Dear Alberto,
>
> On Wed, 23 Feb 2005, Alberto di Bene wrote:
> > Do really exist A/D converters capable of 24-bit resolution at 30
> > MHz bandwidth ?
>
> Why one need 24-bit? Though 16 bit is adequate for most SW,MW,LW
> recievers. 16 bit yelds abt 90 dB dynamic range (every bit exept
> sign bit yelds 6dB).
This means that 1 bit give a sufficient MDS? The recovered audio is
ON/OFF tone?
> But this is dynamic range if there is no
> filtering. When you make banwidth narrow by DSP then dynamic range
> will be improved because noise decreases. If frontend sampling rate
> is, say 100 MHz (there is such an ADC) and bandwidth of DSP filter
> is about, say 10 kHz this yelds 40 dB of noise decreasing. Thus RX
> dynamic range to be about 90+40=130 dB! Realy DD is not 130 dB of
> cose. But it much less due to analog (!) effects in ADC absolutely
> similary as in conventional analog RX. This is not digital effect!
But without a single channel roofing filter a 130 dB outband signal
make the AD converter overload, despite any sampling rate. Or not?
I have some unanswered question:
1- How many bits are needed for a good audio reproduction, i.e. a S/N
ratio of 10 dB with some voice dynamics? Telephone signals usually
are sampled 8 kHz 8 bits for a telephonic quality (with no
compression). May be that 4 bits are sufficient but 1 bit I don't
think.
2- Those bits are to be taken in account for calculating the dynamic
range? i.e. 16 - 4 = 12 bits =72 dB dynamic range.
73 de Claudio, ik2pii
-
Claudio Pozzi - Happy Linux User - http://www.qsl.net/ik2pii
--
Email.it, the professional e-mail, gratis per te: http://www.email.it/f
Sponsor:
Audio, Video, HI-FI...oltre 2.000 prodotti di alta qualità a prezzi da sogno
solo su Visualdream.it
Clicca qui: http://adv.email.it/cgi-bin/foclick.cgi?mid=2954&d=24-2
|
