MIMO OFDM FFT Symbol timing estimation

Hello

As part of a semester long project, I am performing the hardware
implementation of a 2 x 1 MIMO OFDM link. I am formulating a packet
transmission scheme as well as the receiver implementation as part of
this project.. I ran into the following issue, and wondering if anyone
could offer some advice/references.

If you have Q transmit antennas and 1 receive antenna in a MIMO-OFDM
system, the received signal at the antenna is the super-position of Q
transmitted signal (each signal appropriately filtered by the channel
between each transmitter and the receiver).

If this is the case, then the received signal has Q separate FFT start
window locations (corresponding to the FFT start window of every
transmitted signal). In such a case, how does the receiver recognize
where each window starts prior to performing FFT...

Also as a related question, is the preamble (for performing frequency
synchronization) also typically consist of a space-time block code ?

Any help would be greatly appreciated.
Many thanks
Vikram

On 15 Oct 2005 16:53:00 -0700, cvikram@mac.com wrote:

>As part of a semester long project, I am performing the hardware
>implementation of a 2 x 1 MIMO OFDM link. I am formulating a packet
>transmission scheme as well as the receiver implementation as part of
>this project.. I ran into the following issue, and wondering if anyone
>could offer some advice/references.
>
>If you have Q transmit antennas and 1 receive antenna in a MIMO-OFDM
>system, the received signal at the antenna is the super-position of Q
>transmitted signal (each signal appropriately filtered by the channel
>between each transmitter and the receiver).
>
>If this is the case, then the received signal has Q separate FFT start
>window locations (corresponding to the FFT start window of every
>transmitted signal). In such a case, how does the receiver recognize
>where each window starts prior to performing FFT...
>
>Also as a related question, is the preamble (for performing frequency
>synchronization) also typically consist of a space-time block code ?
>
>Any help would be greatly appreciated.
>Many thanks
>Vikram

How to best architect a system to do all of this is still being hotly
debated in many circles.   So you're right at the edge of where a lot
of the industry is working.

I think I see two  questions, but I don't completely understand the
first one.  Are you asking how to synchronize the receiver to the sum
of the transmitted signals, or are you asking how to separate them?

To your second question; if the channels aren't known and are to be
estimated using the preamble (which is often the case), then it isn't
possible to apply STBC to the preambles.  Since decoding of the STBC
requires not only synchronization but knowledge of the channels, it's
tough to apply an STBC to the preamble.

One approach being looked at is to design the preamble such that each
of the spatial channels can be estimated independently, so that the
STBC can be applied immediately at the end of the preamble once the
system is synchronized and the channels estimated.


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

>On 15 Oct 2005 16:53:00 -0700, cvikram@mac.com wrote:
>
>>As part of a semester long project, I am performing the hardware
>>implementation of a 2 x 1 MIMO OFDM link. I am formulating a packet
>>transmission scheme as well as the receiver implementation as part of
>>this project.. I ran into the following issue, and wondering if anyone
>>could offer some advice/references.
>>
>>If you have Q transmit antennas and 1 receive antenna in a MIMO-OFDM
>>system, the received signal at the antenna is the super-position of Q
>>transmitted signal (each signal appropriately filtered by the channel
>>between each transmitter and the receiver).
>>
>>If this is the case, then the received signal has Q separate FFT start
>>window locations (corresponding to the FFT start window of every
>>transmitted signal). In such a case, how does the receiver recognize
>>where each window starts prior to performing FFT...
>>
>>Also as a related question, is the preamble (for performing frequency
>>synchronization) also typically consist of a space-time block code ?
>>
>>Any help would be greatly appreciated.
>>Many thanks
>>Vikram

Hi vikram
           I am also trying a similar scheme using Almouti scheme.firstly
i doubt whthere Q X 1 really falls under the tag name of MIMO..but again
nomeclature is vague..

About your first question ,you have to transmitt both the symbols
simultaneosly.(for the case of almouti scheme or for any other diversity
system for that matter)  
and if we assume we have CP  longer than the delay spread of both
channels, i think you should not have any problem.you can asuume both
signals are reaching simultaneously. for simulation you can add the
signals pointwise at receiver. just remove CP and take FFT after that.

in summary CIRs takes care of your concern about distance travelled from
two antennas to one reciver antennas.

Again, thats how i understand things..

Regrds
sandeep  
		
This message was sent using the Comp.DSP web interface on
www.DSPRelated.com

Thanks for the replies.

I cannot assume that the signals are arriving at equal time/phase.
I have to perform frequency and FFT symbol window synchronization for
each of the transmitted signals within a single receive signal.
Won't each signal act as an interferer for the other signal to make the
timing estimation more difficult...

My guess is that the preambles should be made orthogonal/shift
orthogonal (as is mentioned in a paper by Stuber), but the spectrum of
the resultant signal is no longer band-limited, as we are sending
pseudorandom quasiorthogonal modulated sequences..

Thanks
Vikram

On 16 Oct 2005 06:57:38 -0700, cvikram@mac.com wrote:

>Thanks for the replies.
>
>I cannot assume that the signals are arriving at equal time/phase.
>I have to perform frequency and FFT symbol window synchronization for
>each of the transmitted signals within a single receive signal.
>Won't each signal act as an interferer for the other signal to make the
>timing estimation more difficult...

Unless the distance between the transmit antennas is huge, the timing
misregistration between the incident signals at the receive antenna
will be much less than the channel delay spread.   In this sense
symbol timing recover shouldn't be much of an issue, I'd think, but I
don't know what application you're addressing.

>My guess is that the preambles should be made orthogonal/shift
>orthogonal (as is mentioned in a paper by Stuber), but the spectrum of
>the resultant signal is no longer band-limited, as we are sending
>pseudorandom quasiorthogonal modulated sequences..

Another method is to make sections of the preamble time orthogonal,
i.e., coordinate the transmitters to each transmit a short section of
preamble in a time-multiplexed fashion.   This provides an easy way to
separate the channel estimates at the receiver.   It might make the
preamble a little bit longer, but that's a system tradeoff.


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

Wow, thats pretty interesting, i.e the fact that the timing
misregistration between the different transmit antennas is ignorable..
actually, It seems intuitive. If that is indeed the case, (i.e
identical symbol timing), then the papers are justified in giving a
single symbol timing instant for the received signal, otherwise, they
are building castles in the air.

My project will have 2.4 Ghz antennas separated by less than a 1m
apart. They will transmit using MIMO-OFDM scheme.  I will have a
receiver which will acquire 1 ms of data, demodulated the MISO OFDM
signal and display a clean constellation.

Once the project is over, I plan to put up the source code freely on a
web-site and cover it with a GNU license...

Hi
   I am implementing COFDM. Can any body tell me how actually The Jakes
wireless channel works...Because the parameters and everything are
deterministic.... Plz let me know why.....

Hi vikram
         i think there is serious misunderstanding on the OFDM itself on
your side...   for diversity scheme you have to take one FFT only... and
use linear combiner at the reciver... to recover individual signals...if
you understand equalisation concept in OFDM it would be much easier for
you to make out how things really fits well

                      about preambles i can say that there are multiple
ways of implementing detection schemes... i would rather prefer sending
some pattern from one antenna while sending nothing from other and a
distinct pattern from other antenna while  switching off the other one...
 and dont confuse things with what some of the papers might be saying.. 

you can refer almouti paper for details...

i hope this will help you a lot

sandeep

                                  


>Thanks for the replies.
>
>I cannot assume that the signals are arriving at equal time/phase.
>I have to perform frequency and FFT symbol window synchronization for
>each of the transmitted signals within a single receive signal.
>Won't each signal act as an interferer for the other signal to make the
>timing estimation more difficult...
>
>My guess is that the preambles should be made orthogonal/shift
>orthogonal (as is mentioned in a paper by Stuber), but the spectrum of
>the resultant signal is no longer band-limited, as we are sending
>pseudorandom quasiorthogonal modulated sequences..
>
>Thanks
>Vikram
>
>


		
This message was sent using the Comp.DSP web interface on
www.DSPRelated.com

>My guess is that the preambles should be made orthogonal/shift
>orthogonal (as is mentioned in a paper by Stuber), but the spectrum of
>the resultant signal is no longer band-limited, as we are sending
>pseudorandom quasiorthogonal modulated sequences..

What do you mean with "pseudorandom quasiorthogonal"? 

In the sense of a minimum estimation variance with a least squares
estimator 
the optimal time domain preamble sequences that you send from different TX
antennas are only defined within one OFDM symbol duration. As you read in
those papers, they should have two properties
1. cyclic cross-correlation vanishes within a window whose length is the
guard period.
2. cyclic auto-correlation is perfect within a window whose length is the
guard period.

Taking DFT of those time domain sequences you get pilot sequences defined
in the frequency domain. Each pilot symbol modulates a subcarrier in one
of the Tx antennas and the orthogonality is not destoried at all.

Lanbaba


		
This message was sent using the Comp.DSP web interface on
www.DSPRelated.com

> To your second question; if the channels aren't known and are to be
> estimated using the preamble (which is often the case), then it isn't
> possible to apply STBC to the preambles. Since decoding of the STBC
> requires not only synchronization but knowledge of the channels, it's
> tough to apply an STBC to the preamble.


I don't catch you point. 

In Alamouti's scheme (2x2 STBC) your transmitted signal is orthogonal. If
such an orthogonal design is applied to the preamble for each subcarrier,
then the channel estimation is straight forward given synchronization. Or
do you mean the synchronization is too difficult?

Lanbaba

		
This message was sent using the Comp.DSP web interface on
www.DSPRelated.com