Dear Eric, Tim,
Thanks a lot. THis helps a lot
---------------------------------------
Posted through http://www.DSPRelated.com
Reply by Tim Wescott●February 2, 20162016-02-02
On Tue, 02 Feb 2016 07:00:45 -0600, Sharan123 wrote:
>>So a complex-valued channel gain per antenna gets multiplied by the
>>complex-valued baseband signal received at that antenna. From this
>>perspective the phase is really only relevant to model the phase
>>differences between antennas. For a single antenna system a
>>flat-fading channel would only require a scalar amplitude coefficient,
>>but for multiple antennas the relative phases matter.
>
> Dear Eric & Steve,
>
> Thanks. I assume that the complex-valued channel gain per antenna is a
> single value (one point on complex plane) and not a vector by itself for
> a given antenna ...
I'm not sure if it's entirely clear from the discussion so far, but the
underlying process that you're modeling with antenna stuff is that the
signal goes out from the transmitter, bounces off of random stuff "out
there", and then various bits of it hit the antenna at various delays.
For delays that are a good portion of a cycle of RF, but are small
compared to any modulation interval, the best model is a simple phase
shift and gain (or attenuation). This, in turn, is best modeled as
multiplying the signal with a complex gain.
For delays that are a good portion, or more than one, modulation
interval, then the best model is that delay, plus the afore-mentioned
complex gain. This case is going to be most prevalent when you're
dealing with a spread-spectrum service, because the chipping rate will
tend to be a much healthier portion of the carrier frequency than the
underlying bit rate (or analog signal bandwidth).
There's no reason you can't have multiple paths; for how to handle that
do a search on "rake receiver".
> PS: it is possible that these values could change quasi-dynamically
> based on channel conditions and can vary per antenna
If either your antennas or the things that your signal is reflecting off
of are moving, then yes. So definitely yes if the transmitter or
receiver is mobile (like a cell phone). Buildings, walls, and towers
don't generally get up and walk around, so varying multipath isn't as
much of an issue with fixed stations -- but in a fringe reception area
you can get significant propagation from airplanes and other moving
phenomenon.
--
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
Reply by Eric Jacobsen●February 2, 20162016-02-02
On Tue, 02 Feb 2016 07:00:45 -0600, "Sharan123" <99077@DSPRelated>
wrote:
>>So a complex-valued channel gain per antenna gets multiplied by the
>>complex-valued baseband signal received at that antenna. From this
>>perspective the phase is really only relevant to model the phase
>>differences between antennas. For a single antenna system a
>>flat-fading channel would only require a scalar amplitude coefficient,
>>but for multiple antennas the relative phases matter.
>
>Dear Eric & Steve,
>
>Thanks. I assume that the complex-valued channel gain per antenna is a
>single value (one point on complex plane) and not a vector by itself for a
>given antenna ...
Actually, it can be either. Generally, basic MIMO processing only
needs independent flat-fading (i.e., single value) channel
coefficients to work. In practice, there is often multipath where
the channel impulse response has a delay spread that is represented by
a vector rather than a single coefficient.
>PS: it is possible that these values could change quasi-dynamically based
>on channel conditions and can vary per antenna
Yes. For MIMO to work the individual channels coefficients for each
antenna must be independent of each other to a high degree. That
isn't always the case in practice, but often it is.
Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com
Reply by Sharan123●February 2, 20162016-02-02
>So a complex-valued channel gain per antenna gets multiplied by the
>complex-valued baseband signal received at that antenna. From this
>perspective the phase is really only relevant to model the phase
>differences between antennas. For a single antenna system a
>flat-fading channel would only require a scalar amplitude coefficient,
>but for multiple antennas the relative phases matter.
Dear Eric & Steve,
Thanks. I assume that the complex-valued channel gain per antenna is a
single value (one point on complex plane) and not a vector by itself for a
given antenna ...
PS: it is possible that these values could change quasi-dynamically based
on channel conditions and can vary per antenna
---------------------------------------
Posted through http://www.DSPRelated.com
Reply by Eric Jacobsen●February 1, 20162016-02-01
On Mon, 01 Feb 2016 09:10:42 -0600, "Sharan123" <99077@DSPRelated>
wrote:
>Hello all,
>
>Some context behind my question - this is related to receiver stages in
>wireless communication. In many algorithms related to multiple receiver
>antennae there is a reference to channel gain, which is complex valued. It
>talks about multiplying received signal with complex channel gain
>factors.
>
>I am trying to understand what happens when this is done. I guess it
>alters both amplitude and phase of the signal. Also, after the
>multiplication, how is the signal converted back into real domain from
>complex domain?
>
>PS: I did not want to go into too much of background, as I was trying to
>understand fundamentally what happens if a signal is multiplied by a
>complex number
This is generally modelled at baseband while the signal also has
complex-valued samples. It is possible to do it at an IF while the
signal is real-valued, but it is far simpler to do at baseband with
complex-valued signal samples.
So a complex-valued channel gain per antenna gets multiplied by the
complex-valued baseband signal received at that antenna. From this
perspective the phase is really only relevant to model the phase
differences between antennas. For a single antenna system a
flat-fading channel would only require a scalar amplitude coefficient,
but for multiple antennas the relative phases matter.
Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com
Reply by Steve Pope●February 1, 20162016-02-01
Sharan123 <99077@DSPRelated> wrote:
>Some context behind my question - this is related to receiver stages in
>wireless communication. In many algorithms related to multiple receiver
>antennae there is a reference to channel gain, which is complex valued. It
>talks about multiplying received signal with complex channel gain
>factors.
>I am trying to understand what happens when this is done. I guess it
>alters both amplitude and phase of the signal. Also, after the
>multiplication, how is the signal converted back into real domain from
>complex domain?
This stems from the Turin model, which is the basic multipath channel
model from which other more elaborate channel models are derived.
Turin (who passed away a few years back; some of us here will
remember him from UC Berkeley and Teknekron) posited that for a high
enough carrier frequncy, each reflection would arrive at the antenna
with a uniform random phase shift on the interval [0, 2*pi), and thus
the amplitude of this reflection can be represented by a complex
coefficient.
And further, using this, reflections could then be approximated
by applying time delays plus complex multiplies by these coefficients
to the complex baseband signal from the transmitter, with the result
forming the complex baseband signal in the receiver.
So the idea is, not only do you model the RF channel in a
simple manner, you conveniently bypass needing to fully model the
upconversion/downconversoin to/from RF. Of course you need
to add back in impairments that are bypassed (important ones
being PA non-linearity, receiver noise figure, local oscillator
phase noise, frequency offset and doppler).
So to answer your question in the last sentence above: you don't.
You model the whole thing as a complex signal path, and never convert
to or from a real signal.
Hope this helps.
Steve
Reply by Tauno Voipio●February 1, 20162016-02-01
On 1.2.16 17:10, Sharan123 wrote:
> Hello all,
>
> Some context behind my question - this is related to receiver stages in
> wireless communication. In many algorithms related to multiple receiver
> antennae there is a reference to channel gain, which is complex valued. It
> talks about multiplying received signal with complex channel gain
> factors.
>
> I am trying to understand what happens when this is done. I guess it
> alters both amplitude and phase of the signal. Also, after the
> multiplication, how is the signal converted back into real domain from
> complex domain?
>
> PS: I did not want to go into too much of background, as I was trying to
> understand fundamentally what happens if a signal is multiplied by a
> complex number
This means simply that the channel can change both the
amplitude and the phase of the signal.
--
-TV
Reply by Sharan123●February 1, 20162016-02-01
Hello all,
Some context behind my question - this is related to receiver stages in
wireless communication. In many algorithms related to multiple receiver
antennae there is a reference to channel gain, which is complex valued. It
talks about multiplying received signal with complex channel gain
factors.
I am trying to understand what happens when this is done. I guess it
alters both amplitude and phase of the signal. Also, after the
multiplication, how is the signal converted back into real domain from
complex domain?
PS: I did not want to go into too much of background, as I was trying to
understand fundamentally what happens if a signal is multiplied by a
complex number
---------------------------------------
Posted through http://www.DSPRelated.com
Reply by Nobody●January 31, 20162016-01-31
On Sat, 30 Jan 2016 12:28:14 -0600, Sharan123 wrote:
> I would like to know the implication of multiplying a discrete sinusoidal
> signal with a complex number.
Sinusoid or complex exponential? Signal processing theory tends to use the
latter:
e^(i*x) = cos(x) + i*sin(x)
The main advantage of this over a sinusoid is that you can perform a phase
shift by multiplying by a complex constant.
Reply by Eric Jacobsen●January 31, 20162016-01-31
On Sat, 30 Jan 2016 18:36:26 -0800 (PST), radams2000@gmail.com wrote:
>Well you see, sometimes a discrete sinusoid meets a complex number, and the=
>y decide to multiply (this is now legal in Cambridge Massachusetts where I =
>work). Their child is both discrete and complex (yes I am aware that discre=
>te and discreet are not the same).=20
>
>Bob
Is the interaction of indiscreet sinusoids and complex numbers still
illegal, then?
Reminds me a little of that old, elegant prose, Impure Mathematics
that used to circulate regularly decades ago:
(archived here among other places:
https://www-users.cs.york.ac.uk/susan/joke/polly.htm )
Once upon a time (1/t), pretty little Polly Nomial was strolling
across a field of vectors when she came to the edge of a singularly
large matrix.
Now Polly was convergent and her mother had made it an absolute
condition that she must never enter such an array without her brackets
on. Polly, however, who had changed her variables that morning and was
feeling particularly badly behaved, ignored this condition on the
grounds that it was insufficient and made her way in amongst the
complex elements.
Rows and columns enveloped her on all sides. Tangents approached her
surface. She became tensor and tensor. Quite suddenly, three branches
of a hyperbola touched her at a single point. She oscillated
violently, lost all sense of directrix and went completely divergent.
As she reached a turning point she tripped over a square root which
was protruding from the erf and plunged headlong down a steep
gradient. When she was differentiated once more she found herself,
apparently alone, in a non-euclidean space.
She was being watched, however. That smooth operator, Curly Pi, was
lurking inner product. As his eyes devoured her curvilinear
coordinates, a singular expression crossed his face. Was she still
convergent, he wondered. He decided to integrate improperly at once.
Hearing a vulgar function behind her, Polly turned round and saw Curly
Pi approaching with his power series extrapolated. She could see at
once, by his degenerate conic and his dissipative terms, that he was
bent on no good.
"Eureka" she gasped.
"Ho, ho," he said. "What a symmetric little Polynomial you are. I can
see you're bubbling over with secs".
"O Sir," she protested, "keep away from me. I haven't got my brackets
on."
"Calm yourself, my dear," said our suave operator, "your fears are
purely imaginary "
"i, i," she thought, "perhaps he's homogenous then?".
"What order are you," the brute demanded.
"Seventeen," replied Polly.
Curly leered. "I suppose you've never been operated on yet?" he asked.
"Of course not", Polly cried indignantly. "I'm absolutely convergent."
"Come, come," said Curly. "Let's off to a decimal place I know and
I'll take you to the limit."
"Never," gasped Polly.
"Exchlf," he swore, using the vilest oath he knew. His patience was
gone. Coshing her over the coefficient with a log until she was
powerless, Curly removed her discontinuities. He stared at her
significant places and began to smooth her points of inflexion. Poor
Polly. All was up. She felt his hand tending to her asymptotic limit.
Her convergence would soon be gone forever.
There was no mercy, for Curly was a heavyside operator. He integrated
by parts. He integrated by partial fractions. The complex beast even
went all the way around and did a contour integration. What an
indignity. To be multiply connected on her first integration. Curly
went on operating until he was absolutely and completely orthogonal.
When Polly got home that evening, her mother noticed that she had been
truncated in several places. But it was too late to differentiate now.
As the months went by, Polly increased monotonically. Finally she
generated a small but pathological function which left surds all over
the place until she was driven to distraction.
The moral of this sad story is this: If you want to keep your
expressions convergent, never allow them a single degree of freedom.
Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com