"... 1-bit complex samples at 60 megasamples per second ..." ???

Just what is a "1 bit complex sample"?

"The sampler is driven by orbit-prediction software that triggers 
a synchronized acquisition on both 1575.42 MHz and 1278.75 MHz 
using 1-bit complex samples at 60 megasamples per second (about 
60 MHz total bandwidth)."
 
http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-italy/
  2nd paragraph

>Just what is a "1 bit complex sample"?

My guess is two single bit ADCs 
---------------------------------------
Posted through http://www.DSPRelated.com

On Tue, 11 Aug 2015 15:34:23 -0500, Richard Owlett wrote:

> Just what is a "1 bit complex sample"?
> 
> "The sampler is driven by orbit-prediction software that triggers a
> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using 1-bit
> complex samples at 60 megasamples per second (about 60 MHz total
> bandwidth)."
>  
> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
italy/
>   2nd paragraph

This is all assuming, but:

They've got to be using two 1-bit ADCs, and they've got to be doing I/Q 
sampling.  Whether they're downconverting to baseband and then doing the 
1-bit thang, or downconverting to some IF and directing every other 1-bit 
conversion to a decimator -- I dunno.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

On 8/11/2015 4:50 PM, Tim Wescott wrote:
> On Tue, 11 Aug 2015 15:34:23 -0500, Richard Owlett wrote:
>
>> Just what is a "1 bit complex sample"?
>>
>> "The sampler is driven by orbit-prediction software that triggers a
>> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using 1-bit
>> complex samples at 60 megasamples per second (about 60 MHz total
>> bandwidth)."
>>
>> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
> italy/
>>    2nd paragraph
>
> This is all assuming, but:
>
> They've got to be using two 1-bit ADCs, and they've got to be doing I/Q
> sampling.  Whether they're downconverting to baseband and then doing the
> 1-bit thang, or downconverting to some IF and directing every other 1-bit
> conversion to a decimator -- I dunno.

When you talk about I/Q sampling, should I assume that implies mixing 
with a pair of reference sine waves 90 degrees out of phase before the 
ADC sampling?  Is there an advantage to doing this before sampling 
rather than after?

It has been a few days to my DSP activity, but I want to say every 
system I've worked with had one ADC and generated the I/Q streams 
digitally.  Or is that what you are referring to by "directing every 
other 1-bit conversion to a decimator"?  The method I've seen uses every 
other sample as I and the remaining samples as Q with every other one in 
each stream negated yielding a multiplication by a stream that is ...1, 
0, -1, 0,...

-- 

Rick

On Tue, 11 Aug 2015 17:25:14 -0400, rickman wrote:

> On 8/11/2015 4:50 PM, Tim Wescott wrote:
>> On Tue, 11 Aug 2015 15:34:23 -0500, Richard Owlett wrote:
>>
>>> Just what is a "1 bit complex sample"?
>>>
>>> "The sampler is driven by orbit-prediction software that triggers a
>>> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using
>>> 1-bit complex samples at 60 megasamples per second (about 60 MHz total
>>> bandwidth)."
>>>
>>> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
>> italy/
>>>    2nd paragraph
>>
>> This is all assuming, but:
>>
>> They've got to be using two 1-bit ADCs, and they've got to be doing I/Q
>> sampling.  Whether they're downconverting to baseband and then doing
>> the 1-bit thang, or downconverting to some IF and directing every other
>> 1-bit conversion to a decimator -- I dunno.
> 
> When you talk about I/Q sampling, should I assume that implies mixing
> with a pair of reference sine waves 90 degrees out of phase before the
> ADC sampling?  Is there an advantage to doing this before sampling
> rather than after?

Yes.  If there's an advantage it's just the reduced sampling rate.  But 
the disadvantages engendered by mismatch between your two ADCs may be 
severe.

> It has been a few days to my DSP activity, but I want to say every
> system I've worked with had one ADC and generated the I/Q streams
> digitally.  Or is that what you are referring to by "directing every
> other 1-bit conversion to a decimator"?  The method I've seen uses every
> other sample as I and the remaining samples as Q with every other one in
> each stream negated yielding a multiplication by a stream that is ...1,
> 0, -1, 0,...

That's what I'm talking about: one ADC, then multiply it by two streams:

[ 1  0 -1  0  1  0 -1 ...]
[ 0  1  0 -1  0  1  0 ...]

Keep in mind that at this point we're hypothesizing WAY ahead of our data!

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Richard Owlett <rowlett@cloud85.net> writes:

> Just what is a "1 bit complex sample"?
>
> "The sampler is driven by orbit-prediction software that triggers a
> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using
> 1-bit complex samples at 60 megasamples per second (about 60 MHz total
> bandwidth)."
>
> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-italy/
>  2nd paragraph

Richard,

I think someone's already answered (Tim?), but complex means from the
complex numbers, so of the form a + i*b, where a and b are from a 1-bit
ADC. So it's really two bits total.

Also note that in complex sampling, Nyquist changes: it's the full Fs
(sample rate) and not Fs/2. Actually, it always was Fs, but in a real
signal half of the bandwidth is redundant and therefore wasted.
-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

On Tue, 11 Aug 2015 19:25:03 -0400, Randy Yates wrote:

> Richard Owlett <rowlett@cloud85.net> writes:
> 
>> Just what is a "1 bit complex sample"?
>>
>> "The sampler is driven by orbit-prediction software that triggers a
>> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using
>> 1-bit complex samples at 60 megasamples per second (about 60 MHz total
>> bandwidth)."
>>
>> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
italy/
>>  2nd paragraph
> 
> Richard,
> 
> I think someone's already answered (Tim?), but complex means from the
> complex numbers, so of the form a + i*b, where a and b are from a 1-bit
> ADC. So it's really two bits total.
> 
> Also note that in complex sampling, Nyquist changes: it's the full Fs
> (sample rate) and not Fs/2. Actually, it always was Fs, but in a real
> signal half of the bandwidth is redundant and therefore wasted.

Or it's the way you parse what Harry Nyquist said.  The Nyquist sampling 
theorem says you need to collect independent samples at twice the 
bandwidth.  It doesn't say you have to collect evenly spaced samples of 
the same thing.  Collecting I and Q at the bandwidth is still collecting 
twice the bandwidth.

-- 
www.wescottdesign.com

Richard Owlett wrote:
> Just what is a "1 bit complex sample"?
>
> "The sampler is driven by orbit-prediction software that triggers a
> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using 1-bit
> complex samples at 60 megasamples per second (about 60 MHz total
> bandwidth)."
>
> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-italy/
>
>   2nd paragraph
>

They're probably sampling I and Q seperately.

-- 
Les Cargill

Tim Wescott <tim@seemywebsite.com> writes:

> On Tue, 11 Aug 2015 19:25:03 -0400, Randy Yates wrote:
>
>> Richard Owlett <rowlett@cloud85.net> writes:
>> 
>>> Just what is a "1 bit complex sample"?
>>>
>>> "The sampler is driven by orbit-prediction software that triggers a
>>> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using
>>> 1-bit complex samples at 60 megasamples per second (about 60 MHz total
>>> bandwidth)."
>>>
>>> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
> italy/
>>>  2nd paragraph
>> 
>> Richard,
>> 
>> I think someone's already answered (Tim?), but complex means from the
>> complex numbers, so of the form a + i*b, where a and b are from a 1-bit
>> ADC. So it's really two bits total.
>> 
>> Also note that in complex sampling, Nyquist changes: it's the full Fs
>> (sample rate) and not Fs/2. Actually, it always was Fs, but in a real
>> signal half of the bandwidth is redundant and therefore wasted.
>
> Or it's the way you parse what Harry Nyquist said.  The Nyquist sampling 
> theorem says you need to collect independent samples at twice the 
> bandwidth.  It doesn't say you have to collect evenly spaced samples of 
> the same thing.  Collecting I and Q at the bandwidth is still collecting 
> twice the bandwidth.

I was just going to agree that it depends on perspective, but I just
realized it doesn't. 

Consider sampling I and Q but where Q is zero. In other words, you're
sampling a real signal with a complex ADC. You still have twice as many
samples, but suddenly your (usable) bandwidth is half.

So it's not the samples, it's something more inherent in the source
information.

Resist the temptation to blow this off and say, "Well, setting the
samples to zero is like not sampling." The point is, you are sampling
those imaginary samples, but there's no information there. It's the
source information that is lacking, not the samples.

Or maybe I'm just getting sleepy...
-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

On Tue, 11 Aug 2015 23:35:28 -0400, Randy Yates
<yates@digitalsignallabs.com> wrote:

>Tim Wescott <tim@seemywebsite.com> writes:
>
>> On Tue, 11 Aug 2015 19:25:03 -0400, Randy Yates wrote:
>>
>>> Richard Owlett <rowlett@cloud85.net> writes:
>>> 
>>>> Just what is a "1 bit complex sample"?
>>>>
>>>> "The sampler is driven by orbit-prediction software that triggers a
>>>> synchronized acquisition on both 1575.42 MHz and 1278.75 MHz using
>>>> 1-bit complex samples at 60 megasamples per second (about 60 MHz total
>>>> bandwidth)."
>>>>
>>>> http://gpsworld.com/first-signals-of-beidou-phase-3-acquired-at-ispra-
>> italy/
>>>>  2nd paragraph
>>> 
>>> Richard,
>>> 
>>> I think someone's already answered (Tim?), but complex means from the
>>> complex numbers, so of the form a + i*b, where a and b are from a 1-bit
>>> ADC. So it's really two bits total.
>>> 
>>> Also note that in complex sampling, Nyquist changes: it's the full Fs
>>> (sample rate) and not Fs/2. Actually, it always was Fs, but in a real
>>> signal half of the bandwidth is redundant and therefore wasted.
>>
>> Or it's the way you parse what Harry Nyquist said.  The Nyquist sampling 
>> theorem says you need to collect independent samples at twice the 
>> bandwidth.  It doesn't say you have to collect evenly spaced samples of 
>> the same thing.  Collecting I and Q at the bandwidth is still collecting 
>> twice the bandwidth.
>
>I was just going to agree that it depends on perspective, but I just
>realized it doesn't. 
>
>Consider sampling I and Q but where Q is zero. In other words, you're
>sampling a real signal with a complex ADC. You still have twice as many
>samples, but suddenly your (usable) bandwidth is half.
>
>So it's not the samples, it's something more inherent in the source
>information.
>
>Resist the temptation to blow this off and say, "Well, setting the
>samples to zero is like not sampling." The point is, you are sampling
>those imaginary samples, but there's no information there. It's the
>source information that is lacking, not the samples.
>
>Or maybe I'm just getting sleepy...
>-- 
>Randy Yates
>Digital Signal Labs
>http://www.digitalsignallabs.com

Randy, what they're describing is a common technique where a digital
mixing oscillator mixes the sampled bandwidth to baseband.   The
effect is that fs/4 gets mixed to 0, so the complex mixing phasor has
four samples per cycle.   The phases of the sampling instants are
often chosen to be on the coordinate axes, so that they come out just
like Tim described earlier:

>That's what I'm talking about: one ADC, then multiply it by two streams:

>[ 1  0 -1  0  1  0 -1 ...]
>[ 0  1  0 -1  0  1  0 ...]

Those two vectors are just the I and Q mixing coefficients, in other
words, the mixer samples are just:

1+j0,  0+j1, -1+j0, 0-j1...repeating ad nuaseum...

That's just four samples around the unit circle, which mixes fs/4 to
0.

One consequence of that method is that the information samples don't
line up nicely, so some decimation or other filtering needs to be done
to get things smoothed out.   The decimation process is aided by this
since you know that every other sample in I and Q is zero, so those
calculations can be skipped with no consequence as long as the two
filters are done properly.   Easy decimation by 2:1 and filtering can
be done this way pretty efficiently.

If you don't want to decimate by two for whatever reason, or want to
do it a different way, you can also make things simple by rotating the
mixing coefficients 45 degrees:

1+j1,  -1+j1, -1-j1, 1-j1,  ....forever and ever...

The result is the same, except now you have to process all the
samples, but you also get some good signal magnitude gain because the
vectors are no longer on the unit circle, but have magnitude sqrt(2).
This can be very useful in some circumstances.

Note that all of the information is preserved in either case.   There
is no loss of information or gain in information in either case.   So
those zeros in the first case don't really hurt you at all, and they
can help reduce the computational load signifcantly.

I've built many systems both ways, just depending on what was needed
in each case.   Naturally there are many variations on the theme,
limited only by ones imagination (or schedule).

I just wanted to point out that there's really nothing bad about the
zeros.   The real-valued input stream can be identical in each case,
and you can get the same results in either case just with different
methodology and tradeoffs.


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com