On 4/3/17 9:32 PM, Logan Cummings wrote:

So those folks were trying to use 1 ADC for all three bands, so they had
to choose a sampling rate that lets them separate the signals later in
software.

But that ADC is a MAX104 - a 1GSPS, 8 bit converter - that draws 5 Watts!!!

I'm not sure that's a good trade against a 1 or 2 bit converter for each
band, in terms of the downstream data rate and processing.

Perhaps someone could start a new thread with a different title since this
has evolved to something else then my request.
By the way the response to my question was answered by the group and its
true the first LO does not need to be locked for at least code tracking. I
really didn't prove its needed for carrier tracking.
Thanks everyone.
Paul
WB8TSL

On Tue, Apr 4, 2017 at 9:55 AM, jimlux jimlux@earthlink.net wrote:

On Tue, 4 Apr 2017 06:55:24 -0700
jimlux jimlux@earthlink.net wrote:

Honestly, I don't think the direct sampling approach is a good idea.
It folds a lot of noise into the signal band. I'd rather use a single
heterodyne with an LO frequncy of around 1000MHz, or something between
L2 and E5, such that the bands stay still seperated. Here I would add
a filterbank to get rid of as much noise as possible. And after that
use an ADC sampling frequency to fold the signals down again.
(Effectively forming a super-heterodyne receiver)

You don't need a 1Gsps ADC for that, but if you want to keep all
frequency bands completely seperate, even after sampling, a relatively
high sampling rate is necessary. L1C/E1OS needs at least 14MHz,
L2C needs 2MHz, E5 needs 50MHz. Ie to keep them separated, at least
66MHz of (un-aliased) bandwidth or a sample rate of 132Msps is needed
(alternatively, an I/Q ADC with 66Msps). There are plenty of ADCs
that go up to 120Msps with 10-14bits resolution available, and a couple
that go higher (up to 200MHz are easy to find). 8bit ADCs with >100Msps
are available, but not so many with >120Msps. So, it should be "easy"
to build such a system, if one can find a nice pair of LO frequency
and sampling rate. Alternatively, if one can accept a slightly decreased
SNR one can choose a pair of frequencies where all the signals fall ontop
of eachother, making the 50MHz of E5 the only real requriement. My guess
would be that the CDMA character of the codes would make them easily
seperatable, resulting in an (additional) SNR los of maybe 1-3dB, which
can be compensated by using an ADC with 8bit or even 12bit instead of
the 2-4bit that are now common for GPS receivers. The frequency requirement
can further reduced if one drops the E5b signal and just works with the E5a.
Then 21MHz of bandwidth would be enough.

Looking at the frequency band values, an LO frequency between
1367MHz (L1C touches E5) and 1405MHz (L1C touches L2) would be
the most sensible range. The IF would be below 240MHz and it
seems like the maximum needed bandwidth would be around 70MHz
(eyeballing the graph, no real calculation). I'd say the best
compromise would be using an LO of 1398MHz (IF=170-230MHz)
and using a sampling rate between 120Msps and 160Msps.

The advantage of such a system would be that there is only a single
path through the system for all signals, especially through the filters.
Thus the variability of the differential phase shift between the
frequency bands would be significantly reduced, which would result
in better stability. Of course, that's the theory. Whether things work
out this way in reality is a different question. What can be said for
sure is, because of the high IF frequency of >200MHz, the standard tuner
chips cannot be used anymore and the RX chain has to be build from
"discrete" components, which increases the BOM cost quite considerably.

		Attila Kinali

--
You know, the very powerful and the very stupid have one thing in common.
They don't alters their views to fit the facts, they alter the facts to
fit the views, which can be uncomfortable if you happen to be one of the
facts that needs altering.  -- The Doctor

On 4/4/17 4:21 PM, Attila Kinali wrote:

in most of the designs, the noise is from the LNA, and is band limited,
so the additional noise from the amplifier chain is less. COnsider if
the LNA has 40dB gain and a 2 dB NF. Let's say all the other amps in the
chain have 5 dB NF.

The thermal noise into the next amp is -132 dBm/Hz.  In order for the
5dB NF noise (-169 dBm/Hz) to get up high enough to be noticeable, say,
30dB, you'd have to fold 1000 times the sampling bandwidth. if the
sampling bandwidth is 40 MHz, to get the noise up high enough it would
have to extend to 40 GHz... I'll bet it doesn't<grin>

I don't think keeping the bands together buys you much - you don't need
a multibit ADC for a signal that is below the noise floor. (unless
you're trying to reject strong interference signals, but that's a
different kind of receiver).

Oh, I'm not sure about that. It would depend on the filter kind and
topology.

If it's a SAW or BAW filter, it's all one "brick", but I think you'd
still need to calibrate the differential phase shift vs temp.  And it
might be very predictable in a "measure 10 of them, and now you know the
characteristics of the next 1000"

Of course, that's the theory. Whether things work

There's a ton of integrated demodulator/ADC parts out there these days
that go up to 6GHz.
AD9361 for example

it will do 56 MHz BW through the IF, with 12 bit ADC feeding a 128 tap
FIR filter, etc.

which increases the BOM cost quite considerably.

On 04/05/2017 01:21 AM, Attila Kinali wrote:

Regardless you already have SAW filters on the LNA to provide selectivity.

Also, you don't really need to keep the bands fully separate in their
mixed-down form, since they do not correlate except for the P(Y), but
keeping enough frequency difference, such that doppler shift does not
remove correlation margin, they remain uncorrelated. Some of the
literature pay much attention to the band not wrapping around the
band-edge, but I'm not convinced it is such a big issue.

A direct sampler of 100 MHz would work well for GPS for instance, but
not for GLONASS, but 90 MHz would work there. The S/H would need to have
the BW of the top frequency, but then the S/H action will act as the
first mixer.

Since you don't really need to keep signals very separated, you can pack
them relatively tight. It's the E5 of GALILEO which is wide.

Using a 1,4 GHz range LO1 to pick L1 and L2 has been known to be used
before. There is even existing chips which uses 1.4 GHz on LO1, which
with a different set of filters could almost support L2, will have to
check the details. While that front-end would be neat, I would not use
that chip since it is no longer in production.

The fun thing about these types of receivers, is that there is so many
ways to do it, that it allows for many different approaches to be tried
as technology develops. There is no single one "right" way of doing it.

Cheers,
Magnus

Hi

Galileo E5 is a bit of a strange case. It’s really E5a and E5b. You can either grab it all as one
giant signal or as two separate signals. You may (or may not) care about the data on E5a or
b depending on what you are trying to do. Getting the entire very wide signal likely has some
interesting benefits when it comes to working out very small differences in location or … errr…
time.

One way to do the E5 signal would be a dual (duplicate) IF ISB downconverter. How practical that turns out
to be is an open question. The more conventional approach is to take a monstrous chunk of
L band down to a high speed sampler.

Bob

Hi,

There are many things to be done before attempting the full E5 approach
anyway, so I would not have make it a make or break for a first design.

Cheers,
Magnus

On 04/05/2017 02:27 PM, Bob kb8tq wrote:

On Tue, 4 Apr 2017 17:58:11 -0700
jimlux jimlux@earthlink.net wrote:

The beauty of the system would be that you don't need a SAW filter
at all. If the input stage (LNA + mixer) has a high enough dynamic
range, then the (first) IF filer alone can remove all those out of
band interference. And at the same time, because the IF frequency
being low, you don't need any specialized filter components that
might not be available in a couple of months.

Of course, this doesn't really work that way when significantly
wider signals (E5) have to be caught together with "narrow band"
signals (L1 C/A or L2C).

Unfortunately, the AD9361 does not offer the IF bandwith necessary.
Even though it has a high sample rate and can offer high bandwidth
capture of signals, the zero-IF nature of its design doesn't work
for this design approach. The IF of the AD9361 has a low pass filter
of at most 56MHz, ie it offers to capture a bandwith of 56MHz of
frequency space (using both I and Q channels). But the above approach
would need an IF of >200MHz, but it would be enough to only have a
single channel.

I looked up the prices for the components and figured that the prices for
mixer and IF amplifiers are actually quite low (a 2-4 USD per IC) so it
isn't that much more expensive to build such a system than using a 3 tuner
approach (eg using MAX2120 as Peter Monta did with the GNSS Firehose).

		Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neil Stephenson

On Wed, 5 Apr 2017 10:37:07 +0200
Magnus Danielson magnus@rubidium.dyndns.org wrote:

If part of the signal wraps because you are at the bandedge,
then you lose this part of the signal and the part it wraps over.
This is due to the signal coherently overlapping in frequency space.
As far as I understood the math, there isn't a way to seperate them
again (at least there isn't any I am aware of). Thus this signal energy
is lost for the decoding process.

		Attila Kinali

--
It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neil Stephenson

On Wed, 5 Apr 2017 08:27:58 -0400
Bob kb8tq kb8tq@n1k.org wrote:

I wouldn't call it strange, but rather neat :-)
The E5 signal is created as a single, 8-PSK signal(see [1]), which is
modulated such, that the positive and negative frequency parts get
a specific signal structure. This is done in order to allow an extremely
wide band signal to be demodulated in parts. I guess they feared that a
receiver for a 50MHz wide signal would be too expensive for the
commercial market and made it possible to process the signal as two
20MHz wide pieces. There is a slight loss in correlation energy in this
case, but for most applications it should not matter. The bigger issue
is that the path delays for the two receiver channels would need to be
calibrated and tracked during operation in order to make full use of
the E5 signal.

BTW: I have been told, that using the full E5 signal makes the use
of any other signal kind of unnecessary as its extremely wide bandwidth
allows a very fine tracking of the signal. Thus the use of any other signal
(e.g. E1 OS) would actually degrade the receivers timing performance than
improve it.

As I have written above, to be able to do this is the reason for the E5's
signal structure. And apparently the designers thought that this would be
the way how most users would decode it. I am currently not aware of any
commercial E5 receiver that is already on the market, so it is kind of moot
to ask what the common way to decode E5 is.

BTW: Rodriguez' PhD thesis[2] (which is the basis of navipedia) gives a very
nice overview of the trade-off's that went into the Galileo signals and
gives a few hints where future GNSS signals could further improve things.

		Attila Kinali

[1] Galileo OS SIS ICD Issue 1 Revision 2,
Section 2.3.1.3 "Equivalent Modulation Type"

[2] "On Generalized Signal Waveforms for Satellite Navigation",
by José Ángel Ávila Rodríguez, 2008
https://athene-forschung.unibw.de/node?id=86167

It is upon moral qualities that a society is ultimately founded. All
the prosperity and technological sophistication in the world is of no
use without that foundation.
-- Miss Matheson, The Diamond Age, Neil Stephenson