--- In [email protected], "Patricia Wilson"
<[EMAIL PROTECTED]> wrote:
>
> Thank you Carlos, a useful answer.
> 
> I am rather new to this SDR thing and have a really newby question.
> Where in the chain between antenna and speaker should the signal be
> sampled by an SDR? The Nyquist limit imposes a sample rate of at
> least twice the highest frequency component so if you want to sample
> at the input RF and that RF is 29 MHz then you need a sample rate of
> 58 MHz which is not only pretty fast but will also produce a LOT of
> data very quickly. On the other hand if you mix it down to a first
> IF you lose the possibility of digitally filtering out unwanted
> signals earlier in the signal path before they have a chance to
> overload something.
> 
> As with nearly everything in engineering it is a trade-off. But
> where do most SDR's do the sampling?

 Patricia,

  I am Alberto, not Carlos.

There are two main courses in the SDR schools of thought, each
addressing different needs. The simpler, less expensive, though giving
good results method is that called "the Tayloe mixer". 
The name Tayloe is a bit misleading, as he did not invent that mixer
architecture, but despite prior art he was able to obtain a patent
from the US Patent Office, notorious for its sloppiness...

A SDR based on that mixer architecture, more aptly called QSD
(Quadrature Sampling Detector) uses a local oscillator with two
outputs at 90 degrees from each other that feed the mixer. The LO
frequency is equal to the center frequency of the band segment to be
received. Practically a zero IF. The output of the mixer is a couple
of signals in baseband, called I and Q, which analytically describe
the band of interest.
Those signals are then sent to the stereo inputs of a good quality
sound card, and digitized by the A2D converters on that board.
The sampling frequency used on the sound card determines (in addition
to the hardware filtering in the mixer) the max bandwidth receivable.
Good sound cards can sample up to 192 kHz, and this means that you
have just a bit less than 192 kHz of bandwidth. Nyquist is still
happy, as you work with an analytic signal (the I/Q pair).
Then the software can use the digitized signals for its tasks, like
demodulation and filtering.

The second school of thought is that that places an A2D converter
directly behind the antenna, working at 66 MHz (the SDR-14 and the
SDR-IQ), or 80 MHz (the Perseus), or higher (the upcoming Qs1r and
Mercury receivers). Of course such a high sampling rate is not
directly manageable by a PC, so on the same board of the ADC you will
find a DDC (Digital Down Converter) either implemented using 
commercial chips (e.g. the AD6620 used on the SDR-14 and SDR-IQ
boards), or an FPGA with a custom code. Depending on what has been
programmed on the FPGA you will obtain a downsampled signal (with the
inherent processing gain caused by the downsampling) that can range
from a few kHz to a few MHz of sampling rate. The downsampling process
is done using a quadrature NCO in the DDC, and the result is again an
I/Q pair. Now the data are already digital, and you neednot / cannot
use the sound card to bring them into the PC. Generaly what is used is
a high speed USB 2.0 port, though the use of the Ethernet port is
emerging, thanks to a new product of RFspace, the SDR-IP.

While a sampling rate of 1 MHz can look high, modern dual core CPUs
can cope quite easily with such a high data rate. As an example, I
have an Athlon 64 X2 4500+, and Winrad, when used with Perseus sending
data at 1MHz (that means 6 MB/sec of data, as we have a pair of I/Q
samples, each 24 bits), uses about 35% of the available CPU to do its
tasks of filtering and demodulation. So there is ample romm for higher
sampling rates.

Hope to have explained in clear terms the status of the art about SDR
architectures.    

73  Alberto  I2PHD


Reply via email to