SINR estimation in OFDM systems

HI, here

People usually estimate the SINR in the OFDM systems in the following
way.
1. Estimate over each specific pilot subcarrier of different OFDM
symbols. (i.e. CINR estimation)
2. Then average over all pilot subcarriers (i.e. average CINR) to
approximate the SINR.

The problem is, most CINR estimation algorithms are based on a
simpliied channel model assumption. That is, "y = h*a + w", where,
y denotes the received signal after FFT transformation;
h denotes the equivalent time-varying channel coefficients
a denotes the transmitted signal before IFFT transformation.
w denotes the AWGN noise.
(There is theoretical real SINR for this simplified channel model,
SINR=E{(ha)*(ha)} / E{w*w}; Also, h could be regarded as perfect
channel estimation.)

======

However, I am going with a practical system. That means, I should
justify the SINR estimation method in a complex channel model (not in
a simplified linear version).

Original transmitted signal -> IFFT -> Multipath channel (may result
in ICI here)-> FFT -> Received sigmal.

In this practical channel model.
(It's hard to find theoretical real SINR, also hard to find the
perfect channel estimation, which means it is hard for me to justify
my SINR channel estimation method in the practical systems.)

Could anyone give me some highlight on this problem?

>HI, here
>
>People usually estimate the SINR in the OFDM systems in the following
>way.
>1. Estimate over each specific pilot subcarrier of different OFDM
>symbols. (i.e. CINR estimation)
>2. Then average over all pilot subcarriers (i.e. average CINR) to
>approximate the SINR.
>
>The problem is, most CINR estimation algorithms are based on a
>simpliied channel model assumption. That is, "y = h*a + w", where,
>y denotes the received signal after FFT transformation;
>h denotes the equivalent time-varying channel coefficients
>a denotes the transmitted signal before IFFT transformation.
>w denotes the AWGN noise.
>(There is theoretical real SINR for this simplified channel model,
>SINR=E{(ha)*(ha)} / E{w*w}; Also, h could be regarded as perfect
>channel estimation.)
>
>======
>
>However, I am going with a practical system. That means, I should
>justify the SINR estimation method in a complex channel model (not in
>a simplified linear version).
>
>Original transmitted signal -> IFFT -> Multipath channel (may result
>in ICI here)-> FFT -> Received sigmal.
>
>In this practical channel model.
>(It's hard to find theoretical real SINR, also hard to find the
>perfect channel estimation, which means it is hard for me to justify
>my SINR channel estimation method in the practical systems.)
>
>Could anyone give me some highlight on this problem?
>
>

I would implment the OFDM receiver, demodulate the signal and subtract it
from the received signal to get an estimate of all noise plus interference
corresponding to your specific receiver structure and the real-world
conditions. The calculation of the SINR is then trivial.

LBB



_____________________________________
Do you know a company who employs DSP engineers?  
Is it already listed at http://dsprelated.com/employers.php ?

> I would implment the OFDM receiver, demodulate the signal and subtract it
> from the received signal to get an estimate of all noise plus interference
> corresponding to your specific receiver structure and the real-world
> conditions. The calculation of the SINR is then trivial.

===

Hmmm...

"implmenting the OFDM receiver, demodulate the signal" sounds easy,
but as to "Subtract it from the received signal", hmmm, I can't see
where is the received signal.

Do you mean that

after we get the received signal(say y)  from the channelthen we do
the FFT transfomation, demodulate the signal,

and then modulate the signal again, do the IFFT transformation
again(say the result is x'),

So, we can subtract x' from the received signal y to do the SINR
estimation?

--
Jia

To LBB,

Moreover, the way you said is something like "Non-data-aided" SINR
estimation method.

However, I want to do some "Data-aided" SINR estimatiion method, which
means I know some transmitted pilots beforehand.

=====

And also, the channel is fading channel, not AWGN channel. So, we may
not just subtract the demodulated signal from the received signal
directly. We need to know the channel fading coefficient.

---

Jia

>To LBB,
>
>Moreover, the way you said is something like "Non-data-aided" SINR
>estimation method.
>
>However, I want to do some "Data-aided" SINR estimatiion method, which
>means I know some transmitted pilots beforehand.
>
>=====
>
>And also, the channel is fading channel, not AWGN channel. So, we may
>not just subtract the demodulated signal from the received signal
>directly. We need to know the channel fading coefficient.
>
>---
>
>Jia
>
>

Sorry for the oversimplified answer.

Ususaaly after the time/frequency synchronization, channel coefficient
estimation, one-tap frequency domain equalization you get a series of
decision variables for all subcarriers within one OFDM symbol duration,
based on which you can make symbol detection, demapping, deinterleaving,
decoding and so on. Then you can use the decoded bit stream to reconstruct
the signal on each subcarrier by re-interleaving, re-coding and re-mapping.
This reconstructed signal is supposed to be subtracted from the decision
variable, which leads to the estimate of the noise plus interference. The
assumption here is a quasi-error free reception after the decoding,
meaning you can really extract the transmitted signal by your receiver
under the given transmission condition. Otherwise you should expect a
large deviation of the estimated SINR. But if you intend to use a training
sequence you don't need to reconstruct the signal any more. Instead you can
subtract the training sequence, which is usually defined in the frequency
domain, from the decision variable, and this simplifies the whole story
quite a lot.

Yes, there will be frequency selective fading for multipath channels and
fast/slow fading for mobile channels. By the reconstruction and
subtraction mehtod you can work out an average SINR over time and
frequency. You can also estimate the SINR of a specific subcarrier over
time if you like. As mentioned, the statics become trivial once you
reconstruct the signal. 

LBB

_____________________________________
Do you know a company who employs DSP engineers?  
Is it already listed at http://dsprelated.com/employers.php ?

no, the answer is not correct.
People doesn't concern SINR but CINR, and the definition of CINR is:
CINR= power(H*X)/power(W+I) also, CINR = power(H)/(power(W+I)/X)

we can't just subtract the X from the received signal in frequency domain
to find out the value of (W+I). We should subtract (H*X)which is useful
signal. The difficulty is that H is a variant in a real channel. Some
people study the relation between adjacent subcarriers,and other study the
time domain channel impulse response characteristic.But all of this studies
are not perfect.

If you are interested it,please contact me. 
wanghui263@hotmail.

>>To LBB,
>>
>>Moreover, the way you said is something like "Non-data-aided" SINR
>>estimation method.
>>
>>However, I want to do some "Data-aided" SINR estimatiion method, which
>>means I know some transmitted pilots beforehand.
>>
>>=====
>>
>>And also, the channel is fading channel, not AWGN channel. So, we may
>>not just subtract the demodulated signal from the received signal
>>directly. We need to know the channel fading coefficient.
>>
>>---
>>
>>Jia
>>
>>
>
>Sorry for the oversimplified answer.
>
>Ususaaly after the time/frequency synchronization, channel coefficient
>estimation, one-tap frequency domain equalization you get a series of
>decision variables for all subcarriers within one OFDM symbol duration,
>based on which you can make symbol detection, demapping, deinterleaving,
>decoding and so on. Then you can use the decoded bit stream to
reconstruct
>the signal on each subcarrier by re-interleaving, re-coding and
re-mapping.
>This reconstructed signal is supposed to be subtracted from the decision
>variable, which leads to the estimate of the noise plus interference.
The
>assumption here is a quasi-error free reception after the decoding,
>meaning you can really extract the transmitted signal by your receiver
>under the given transmission condition. Otherwise you should expect a
>large deviation of the estimated SINR. But if you intend to use a
training
>sequence you don't need to reconstruct the signal any more. Instead you
can
>subtract the training sequence, which is usually defined in the
frequency
>domain, from the decision variable, and this simplifies the whole story
>quite a lot.
>
>Yes, there will be frequency selective fading for multipath channels and
>fast/slow fading for mobile channels. By the reconstruction and
>subtraction mehtod you can work out an average SINR over time and
>frequency. You can also estimate the SINR of a specific subcarrier over
>time if you like. As mentioned, the statics become trivial once you
>reconstruct the signal. 
>
>LBB
>
>_____________________________________
>Do you know a company who employs DSP engineers?  
>Is it already listed at http://dsprelated.com/employers.php ?
>