Simulation of Antenna Diversity

Hi,

I want to measure performance gains of some common antenna diversity
schemes as used in Mobile Wireless. Can I just simulate uncorrelated
multipath fading over two antennas and add the faded signals from two
transmit antennas at the receiver ? Do I need to take care of any
other issues.

Sachin

sacgupta@hotmail.com (Sachin Gupta) wrote in message news:<1750dd51.0307260339.1fd10a9a@posting.google.com>...
> Hi,
> 
> I want to measure performance gains of some common antenna diversity
> schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> multipath fading over two antennas and add the faded signals from two
> transmit antennas at the receiver ? Do I need to take care of any
> other issues.
> 
> Sachin

Hi Sachin,

I guess you hinted at MRC gain in two-way diversity configuration. If
you are
interested in static absolute sensitivity and multipath reference
sensitivities of the receiver then your model is fine and MRC will
improve the receiver sensitivity by 2-3dB for static case and 4-5dB
for mutipath case on the average.

You will get sufficient gain for ACI and CCI(Co channel interference)
reference level measurement but not the best performance. You need to
perform IRC (interference Rejection Combining) algorithm which will
give the best performance in your case when both faded signals
received are uncorrelated and independently faded.

Hope this very brief overview  helps .

Santosh

On 26 Jul 2003 04:39:26 -0700, sacgupta@hotmail.com (Sachin Gupta)
wrote:

>Hi,
>
>I want to measure performance gains of some common antenna diversity
>schemes as used in Mobile Wireless. Can I just simulate uncorrelated
>multipath fading over two antennas and add the faded signals from two
>transmit antennas at the receiver ? Do I need to take care of any
>other issues.
>
>Sachin

Sachin,

That's the simplest thing to do, but your results may be skewed as a
result.  Many of the diversity schemes rely on the channels being
uncorrelated, so if you always feed uncorrelated signals into the
receive antennas you'll always get good results.   This isn't always
real-world relevant, though, so a good multipath propagation model is
useful to further sort the wheat from the chaff.

Unfortunately, this is still an area of active research.   A lot of
people are spending a lot of time and money doing data collection on
MIMO propagation (which is also relevant to plain antenna diversity
systems), but there doesn't seem to be total agreement yet on good
models.

Depending on what you're simulating, though, you may be able to get
fairly relevant results with some simple models.   How decorrelated
the receive antenna channels will be depends on how far they're
separated.  It appears that at least one wavelength of separation is
required, and personally I think the more the merrier (i.e., if you
can get a few wavelengths of separation, it's a good thing).

FWIW, many people are doing exactly what you suggest and just
generating indepedent, uncorrelated channels into each input.  My
reaction is then to interpret the results as an upper bound on
performance and not necessarily what you'll get in the real world.


Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

Aloha,

Sachin Gupta schrieb:
> I want to measure performance gains of some common antenna diversity
> schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> multipath fading over two antennas and add the faded signals from two
> transmit antennas at the receiver ? Do I need to take care of any
> other issues.

With two completely independent channels you are simulating only one 
of the cases you may find in reality. Other cases will be partly or allmost
fully correlated channels (from every BS antenna to every MS antenna).
The correlation between the channels is the spatial caracteristic of
the radio channel based n the geometry. 
One of the approaches to transform a geometry to a spatial/temporal
channel model you can find in
3GPP TR25.996, on www.3gpp.org.

Wishing a happy day
		LOBI

eric.jacobsen@ieee.org (Eric Jacobsen) wrote in message news:<3f22c221.49066687@news.earthlink.net>...
> On 26 Jul 2003 04:39:26 -0700, sacgupta@hotmail.com (Sachin Gupta)
> wrote:
> 
> >Hi,
> >
> >I want to measure performance gains of some common antenna diversity
> >schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> >multipath fading over two antennas and add the faded signals from two
> >transmit antennas at the receiver ? Do I need to take care of any
> >other issues.
> >
> >Sachin
> 
> Sachin,
> 
> That's the simplest thing to do, but your results may be skewed as a
> result.  Many of the diversity schemes rely on the channels being
> uncorrelated, so if you always feed uncorrelated signals into the
> receive antennas you'll always get good results.   This isn't always
> real-world relevant, though, so a good multipath propagation model is
> useful to further sort the wheat from the chaff.

When I started this simulation stuff, the references I had told me to
simulate uncorrelated multipaths. So the classical Jakes model which
gives correlated multipath was modified (according to one of IEEE
papers) to one which now gives uncorrelated multipaths (very less
correlation over a long duration). Now you suggest that I start
getting correlated multipaths back ! See more on this below.
 
> Unfortunately, this is still an area of active research.   A lot of
> people are spending a lot of time and money doing data collection on
> MIMO propagation (which is also relevant to plain antenna diversity
> systems), but there doesn't seem to be total agreement yet on good
> models.
> 
> Depending on what you're simulating, though, you may be able to get
> fairly relevant results with some simple models.   How decorrelated
> the receive antenna channels will be depends on how far they're
> separated.  It appears that at least one wavelength of separation is
> required, and personally I think the more the merrier (i.e., if you
> can get a few wavelengths of separation, it's a good thing).
> 
> FWIW, many people are doing exactly what you suggest and just
> generating indepedent, uncorrelated channels into each input.  My
> reaction is then to interpret the results as an upper bound on
> performance and not necessarily what you'll get in the real world.

I may have given the wrong impression that I wanted to measure
performance of Diversity combining schemes. I actually want to get my
simulation model right and then get my diversity combining schemes
right. I am not evaluating/performance benchmarking different
combining schemes - they are predefined.

Do I need to look into Spatial Channel Modelling ? (I tried to take a
look at it but it seems very complicated for my application - the
channel model itself would take more time than the combining schemes)

What about the Modified Jakes' fading model - do I go back to the
original (correlated) Jakes model ?

> Eric Jacobsen
> Minister of Algorithms, Intel Corp.
> My opinions may not be Intel's opinions.
> http://www.ericjacobsen.org

Thanks a lot for your thoughts, they have helped a lot. 

Sachin

Andreas Lobinger <andreas.lobinger@netsurf.de> wrote in message news:<3F24EC17.9700EAF0@netsurf.de>...
> Aloha,
> 
> Sachin Gupta schrieb:
> > I want to measure performance gains of some common antenna diversity
> > schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> > multipath fading over two antennas and add the faded signals from two
> > transmit antennas at the receiver ? Do I need to take care of any
> > other issues.
> 
> With two completely independent channels you are simulating only one 
> of the cases you may find in reality. Other cases will be partly or allmost
> fully correlated channels (from every BS antenna to every MS antenna).
> The correlation between the channels is the spatial caracteristic of
> the radio channel based n the geometry. 
> One of the approaches to transform a geometry to a spatial/temporal
> channel model you can find in
> 3GPP TR25.996, on www.3gpp.org.
> 
> Wishing a happy day
> 		LOBI

LOBI, 

Thanks for the reference. Looks very complicated ! Any tutorial (back
to basics) type paper on this ?

I have one more question, the performance would degrade with
correlated multipaths. But the diversity combining scheme itself
should not change ? (no MUD or interference cancellation being used)

Sachin

Aloha,

Sachin Gupta schrieb:
> Andreas Lobinger <andreas.lobinger@netsurf.de> wrote in message > > Sachin Gupta schrieb:
> > > I want to measure performance gains of some common antenna diversity
> > > schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> > > multipath fading over two antennas and add the faded signals from two
> > > transmit antennas at the receiver ? Do I need to take care of any
> > > other issues.

> > One of the approaches to transform a geometry to a spatial/temporal
> > channel model you can find in
> > 3GPP TR25.996, on www.3gpp.org.

> Thanks for the reference. Looks very complicated ! Any tutorial (back
> to basics) type paper on this ?

Of course the SCM is a very detailed version of a spatial channel
model. But you should see here, that the Jakes (uncorrelated 
omnidirectional multipath) Spectrum is dropped for a more geometric
approach. If you have one omnidirectional antenna at the
MS you could do (as everyone starting with) working with independent
uncorrelated channels and correlate them with the Channel
Covariance Matrix. A view in 25.869 V1.00 can give you a few hints
here.
 
> I have one more question, the performance would degrade with
> correlated multipaths. But the diversity combining scheme itself
> should not change?

Putting it this way: If you have a diversity receiver concept that
depends on uncorrelated channels, you will see a degradation with
increasing correlation. But there are other concepts which can
deal with partially correlated channels or concepts that exploit
the spatial nature of the channel... (MIMO etc.).

Wishing a happy day
		LOBI

On 28 Jul 2003 06:00:54 -0700, sacgupta@hotmail.com (Sachin Gupta)
wrote:

>eric.jacobsen@ieee.org (Eric Jacobsen) wrote in message news:<3f22c221.49066687@news.earthlink.net>...
>> On 26 Jul 2003 04:39:26 -0700, sacgupta@hotmail.com (Sachin Gupta)
>> wrote:
>> 
>> >Hi,
>> >
>> >I want to measure performance gains of some common antenna diversity
>> >schemes as used in Mobile Wireless. Can I just simulate uncorrelated
>> >multipath fading over two antennas and add the faded signals from two
>> >transmit antennas at the receiver ? Do I need to take care of any
>> >other issues.
>> >
>> >Sachin
>> 
>> Sachin,
>> 
>> That's the simplest thing to do, but your results may be skewed as a
>> result.  Many of the diversity schemes rely on the channels being
>> uncorrelated, so if you always feed uncorrelated signals into the
>> receive antennas you'll always get good results.   This isn't always
>> real-world relevant, though, so a good multipath propagation model is
>> useful to further sort the wheat from the chaff.
>
>When I started this simulation stuff, the references I had told me to
>simulate uncorrelated multipaths. So the classical Jakes model which
>gives correlated multipath was modified (according to one of IEEE
>papers) to one which now gives uncorrelated multipaths (very less
>correlation over a long duration). Now you suggest that I start
>getting correlated multipaths back ! See more on this below.

Yeah, sorry.  ;)
 
>> Unfortunately, this is still an area of active research.   A lot of
>> people are spending a lot of time and money doing data collection on
>> MIMO propagation (which is also relevant to plain antenna diversity
>> systems), but there doesn't seem to be total agreement yet on good
>> models.
>> 
>> Depending on what you're simulating, though, you may be able to get
>> fairly relevant results with some simple models.   How decorrelated
>> the receive antenna channels will be depends on how far they're
>> separated.  It appears that at least one wavelength of separation is
>> required, and personally I think the more the merrier (i.e., if you
>> can get a few wavelengths of separation, it's a good thing).
>> 
>> FWIW, many people are doing exactly what you suggest and just
>> generating indepedent, uncorrelated channels into each input.  My
>> reaction is then to interpret the results as an upper bound on
>> performance and not necessarily what you'll get in the real world.
>
>I may have given the wrong impression that I wanted to measure
>performance of Diversity combining schemes. I actually want to get my
>simulation model right and then get my diversity combining schemes
>right. I am not evaluating/performance benchmarking different
>combining schemes - they are predefined.
>
>Do I need to look into Spatial Channel Modelling ? (I tried to take a
>look at it but it seems very complicated for my application - the
>channel model itself would take more time than the combining schemes)
>
>What about the Modified Jakes' fading model - do I go back to the
>original (correlated) Jakes model ?

I think it just depends on what you want to do and how much time you
want to spend on it.   As Lobi suggested you can correlate independant
channels according to some channel covariance matrix, and you may be
able to make some assumptions to generate the covariance matrix (or
just simulate across some representative samples of covariance
matrices).

It is very time consuming to try to "accurately" model a
multi-antenna system since it is so dependant on the assumed
environment, spacing and directivity of the antennas, etc., etc.   No
matter what you assume it will probably cover only a fraction of the
possible cases.  I think this is why a lot of people just punt and
assume uncorrelated channels, and then wave their hands that that can
be assured with adequate antenna separation.

If you're working on a cellular system then the Jakes model is a
reasonable thing to use.  It is well understood and seems to be widely
accepted among the cellular community, although it does have
limitations.  If you're working on some other application, however,
you may want to look at other methods (e.g., Jakes won't be
appropriate for WLAN, sensor networks, etc.).

Cheers,



Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org

eric.jacobsen@ieee.org (Eric Jacobsen) wrote in message news:<3f256466.221706514@news.earthlink.net>...
> On 28 Jul 2003 06:00:54 -0700, sacgupta@hotmail.com (Sachin Gupta)
> wrote:
> 
> >eric.jacobsen@ieee.org (Eric Jacobsen) wrote in message news:<3f22c221.49066687@news.earthlink.net>...
> >> On 26 Jul 2003 04:39:26 -0700, sacgupta@hotmail.com (Sachin Gupta)
> >> wrote:
> >> 
> >> >Hi,
> >> >
> >> >I want to measure performance gains of some common antenna diversity
> >> >schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> >> >multipath fading over two antennas and add the faded signals from two
> >> >transmit antennas at the receiver ? Do I need to take care of any
> >> >other issues.
> >> >
> >> >Sachin
> >> 
> >> Sachin,
> >> 
> >> That's the simplest thing to do, but your results may be skewed as a
> >> result.  Many of the diversity schemes rely on the channels being
> >> uncorrelated, so if you always feed uncorrelated signals into the
> >> receive antennas you'll always get good results.   This isn't always
> >> real-world relevant, though, so a good multipath propagation model is
> >> useful to further sort the wheat from the chaff.
> >
> >When I started this simulation stuff, the references I had told me to
> >simulate uncorrelated multipaths. So the classical Jakes model which
> >gives correlated multipath was modified (according to one of IEEE
> >papers) to one which now gives uncorrelated multipaths (very less
> >correlation over a long duration). Now you suggest that I start
> >getting correlated multipaths back ! See more on this below.
> 
> Yeah, sorry.  ;)
>  
> >> Unfortunately, this is still an area of active research.   A lot of
> >> people are spending a lot of time and money doing data collection on
> >> MIMO propagation (which is also relevant to plain antenna diversity
> >> systems), but there doesn't seem to be total agreement yet on good
> >> models.
> >> 
> >> Depending on what you're simulating, though, you may be able to get
> >> fairly relevant results with some simple models.   How decorrelated
> >> the receive antenna channels will be depends on how far they're
> >> separated.  It appears that at least one wavelength of separation is
> >> required, and personally I think the more the merrier (i.e., if you
> >> can get a few wavelengths of separation, it's a good thing).
> >> 
> >> FWIW, many people are doing exactly what you suggest and just
> >> generating indepedent, uncorrelated channels into each input.  My
> >> reaction is then to interpret the results as an upper bound on
> >> performance and not necessarily what you'll get in the real world.
> >
> >I may have given the wrong impression that I wanted to measure
> >performance of Diversity combining schemes. I actually want to get my
> >simulation model right and then get my diversity combining schemes
> >right. I am not evaluating/performance benchmarking different
> >combining schemes - they are predefined.
> >
> >Do I need to look into Spatial Channel Modelling ? (I tried to take a
> >look at it but it seems very complicated for my application - the
> >channel model itself would take more time than the combining schemes)
> >
> >What about the Modified Jakes' fading model - do I go back to the
> >original (correlated) Jakes model ?
> 
> I think it just depends on what you want to do and how much time you
> want to spend on it.   As Lobi suggested you can correlate independant
> channels according to some channel covariance matrix, and you may be
> able to make some assumptions to generate the covariance matrix (or
> just simulate across some representative samples of covariance
> matrices).
> 
> It is very time consuming to try to "accurately" model a
> multi-antenna system since it is so dependant on the assumed
> environment, spacing and directivity of the antennas, etc., etc.   No
> matter what you assume it will probably cover only a fraction of the
> possible cases.  I think this is why a lot of people just punt and
> assume uncorrelated channels, and then wave their hands that that can
> be assured with adequate antenna separation.
> 
> If you're working on a cellular system then the Jakes model is a
> reasonable thing to use.  It is well understood and seems to be widely
> accepted among the cellular community, although it does have
> limitations.  If you're working on some other application, however,
> you may want to look at other methods (e.g., Jakes won't be
> appropriate for WLAN, sensor networks, etc.).

Eric,

Could you elaborate a bit here why Jakes model won't be appropriate for WLAN?

Santosh

> 
> Cheers,
> 
> 
> 
> Eric Jacobsen
> Minister of Algorithms, Intel Corp.
> My opinions may not be Intel's opinions.
> http://www.ericjacobsen.org

Andreas Lobinger <andreas.lobinger@netsurf.de> wrote in message news:<3F2525CE.DD6B9EE7@netsurf.de>...
> Aloha,
> 
> Sachin Gupta schrieb:
> > Andreas Lobinger <andreas.lobinger@netsurf.de> wrote in message > > Sachin Gupta schrieb:
> > > > I want to measure performance gains of some common antenna diversity
> > > > schemes as used in Mobile Wireless. Can I just simulate uncorrelated
> > > > multipath fading over two antennas and add the faded signals from two
> > > > transmit antennas at the receiver ? Do I need to take care of any
> > > > other issues.
>  
> > > One of the approaches to transform a geometry to a spatial/temporal
> > > channel model you can find in
> > > 3GPP TR25.996, on www.3gpp.org.
>  
> > Thanks for the reference. Looks very complicated ! Any tutorial (back
> > to basics) type paper on this ?
> 
> Of course the SCM is a very detailed version of a spatial channel
> model. But you should see here, that the Jakes (uncorrelated 
> omnidirectional multipath) Spectrum is dropped for a more geometric
> approach. If you have one omnidirectional antenna at the
> MS you could do (as everyone starting with) working with independent
> uncorrelated channels and correlate them with the Channel
> Covariance Matrix. A view in 25.869 V1.00 can give you a few hints
> here.

Thanks for this reference. I'll try to look into it also and see what
I can do.

> > I have one more question, the performance would degrade with
> > correlated multipaths. But the diversity combining scheme itself
> > should not change?
> 
> Putting it this way: If you have a diversity receiver concept that
> depends on uncorrelated channels, you will see a degradation with
> increasing correlation. But there are other concepts which can
> deal with partially correlated channels or concepts that exploit
> the spatial nature of the channel... (MIMO etc.).

Thanks for this thought also. As I wrote I was under impression that
nothing much could be done about correlation with TX antenna diversity
alone. I'll try this also.

> Wishing a happy day
> 		LOBI

Wishing the same for you. Your comments have been most helpful. 
Sachin