Channel Estimation During Data period

Hi All



I am simulating time varying freq selective channel where the channel is
changing for every symbol. The packet format is 511 training sequence
followed by 2000 data symbols.

Now at the receiver, as the training sequence is known, I am able to
estimate the channel perfectly but is there anyway to estimate the channel
during data period?

Thanks

Best Regards,

Chintan

On 5 Aug, 12:42, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hi All
>
> I am simulating time varying freq selective channel where the channel is
> changing for every symbol. The packet format is 511 training sequence
> followed by 2000 data symbols.
>
> Now at the receiver, as the training sequence is known, I am able to
> estimate the channel perfectly but is there anyway to estimate the channel
> during data period?

The simple answer is 'no.' Since you don't know what was
sent, you don't know what you ought to recieve, so there
is no way to correct for transmission errors.

The not-quite-so-simple answer is 'maybe.' *If* you trust
the decoder enough to assume that it came up with the
correct symbols, you could go back and analyze the recieved
waveform and use knowledge about the decoded symbol as a
template for what the transmitter ought to have sent.

This *might* work, *if* the channel distorsion is 'small.'
However, if the decoder somehow produces some other symbol
than was actually sent, you have trouble. In that case
you might experience a switch from a poor channel model
with associated poor (but working) communication, to a
totally wrong channel model with nonexistant communication.

Rune

>On 5 Aug, 12:42, "cpshah99" <cpsha...@rediffmail.com> wrote:
>> Hi All
>>
>> I am simulating time varying freq selective channel where the channel
is
>> changing for every symbol. The packet format is 511 training sequence
>> followed by 2000 data symbols.
>>
>> Now at the receiver, as the training sequence is known, I am able to
>> estimate the channel perfectly but is there anyway to estimate the
channel
>> during data period?
>
>The simple answer is 'no.' Since you don't know what was
>sent, you don't know what you ought to recieve, so there
>is no way to correct for transmission errors.
>
>The not-quite-so-simple answer is 'maybe.' *If* you trust
>the decoder enough to assume that it came up with the
>correct symbols, you could go back and analyze the recieved
>waveform and use knowledge about the decoded symbol as a
>template for what the transmitter ought to have sent.
>
>This *might* work, *if* the channel distorsion is 'small.'
>However, if the decoder somehow produces some other symbol
>than was actually sent, you have trouble. In that case
>you might experience a switch from a poor channel model
>with associated poor (but working) communication, to a
>totally wrong channel model with nonexistant communication.
>
>Rune
>

Hi Rune,

I thought so. Actually, I did soem experiment and I wanted to do channel
estimation. From the channel estimation that I did, it turns out that it
was pretty nasty channel.

And for that channel my receiver gives 0.3 BER.

Just one more que: if the channel is completely characterised as Rayleigh,
i.e. time varying freq selective Rayleigh fading, then is it true that
adaptive DFE will fail?

Thanks again

Chintan

Google "blind equalization".
STUPIDENT


cpshah99 wrote:
> Hi All
> 
> 
> 
> I am simulating time varying freq selective channel where the channel is
> changing for every symbol. The packet format is 511 training sequence
> followed by 2000 data symbols.
> 
> Now at the receiver, as the training sequence is known, I am able to
> estimate the channel perfectly but is there anyway to estimate the channel
> during data period?
> 
> Thanks
> 
> Best Regards,
> 
> Chintan

>
>Google "blind equalization".
>STUPIDENT
>
>

Hi Vlad.

I know about blind equalization but I have long training sequence. So
there is no point for blind equalization. Sure I can treat training as data
but it is also true that blind equalizers do not perform as good as the
ones with training. 

Besides I am confident that blind equalization will not do any good in my
case.

Thanks

Chintan

cpshah99 wrote:

>>Google "blind equalization".
>>STUPIDENT
>>
>>
> Hi Vlad.
> 
> I know about blind equalization but I have long training sequence. So
> there is no point for blind equalization. Sure I can treat training as data
> but it is also true that blind equalizers do not perform as good as the
> ones with training. 

Do as all modems do: when receiving the data, update the equalizer based 
on the error vector between the expected symbol and the actually 
received symbol.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

On 8/5/2009 3:42 AM, cpshah99 wrote:
> Hi All
>
>
>
> I am simulating time varying freq selective channel where the channel is
> changing for every symbol. The packet format is 511 training sequence
> followed by 2000 data symbols.
>
> Now at the receiver, as the training sequence is known, I am able to
> estimate the channel perfectly but is there anyway to estimate the channel
> during data period?
>
> Thanks
>
> Best Regards,
>
> Chintan

As VV has hinted, it is possible to update the channel estimate as you 
go along, but if the channel changes very much from symbol to symbol it 
might not be practical.

In OFDM systems the use of pilot tones helps for updating the channel 
estimate on the fly.  If one were to design a system from scratch to 
handle such fast channel changes one approach would be to have about as 
many pilot tones as data tones.  Likewise in a single carrier system a 
repeating training sequence or pilot symbols will help.

When the channel coherence time gets so small that the channel changes 
substantially from symbol to symbol, generally all bets are off.  That's 
a nearly impossible task unless you're willing to spend a LOT of power 
and bandwidth overhead on pilot tones and/or other mechanisms to 
facilitate faster adaptation.

Cheers,
-- 
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.abineau.com

>
>As VV has hinted, it is possible to update the channel estimate as you 
>go along, but if the channel changes very much from symbol to symbol it 
>might not be practical.
>
>In OFDM systems the use of pilot tones helps for updating the channel 
>estimate on the fly.  If one were to design a system from scratch to 
>handle such fast channel changes one approach would be to have about as 
>many pilot tones as data tones.  Likewise in a single carrier system a 
>repeating training sequence or pilot symbols will help.
>
>When the channel coherence time gets so small that the channel changes 
>substantially from symbol to symbol, generally all bets are off.  That's

>a nearly impossible task unless you're willing to spend a LOT of power 
>and bandwidth overhead on pilot tones and/or other mechanisms to 
>facilitate faster adaptation.
>
>

Hi Eric

I agree with you about OFDM. But when I did the channel estimation during
training, I could see that channel is about every 10 to 20 symbols. 

I have seen some papers where they distribute training seq peridically.

But in my case, I would rather use ARQ protocol. :-)

Thanks everybody.

Chintan

On 5 Aug, 19:52, Eric Jacobsen <eric.jacob...@ieee.org> wrote:

> When the channel coherence time gets so small that the channel changes
> substantially from symbol to symbol, generally all bets are off.

This is the reason why underwater communication is hard.
The channel characteristics can change substantially over
a small area, and over a short time.

Within a couple of meters, you can loose or gain yet another
multipath. Within minutes, the tide can change the depth
enough to signifcicantly change the multipath pattern.

>&#4294967295;That's
> a nearly impossible task unless you're willing to spend a LOT of power
> and bandwidth overhead on pilot tones and/or other mechanisms to
> facilitate faster adaptation.

Under water, you con't have much bandwidth. First of all,
absorption losses are proportional to frequency. Scattering
also becomes worse at high frequencies, so you want to
work in low frequency bands. But then geometric effects
come in to play, with normal mode interference from
both sea surface and the bottom.

And of course, the andwidth you do have, is contamicated
by more or less random interference caused by severe
multipath propagation.

Rune

>On 5 Aug, 19:52, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
>
>> When the channel coherence time gets so small that the channel changes
>> substantially from symbol to symbol, generally all bets are off.
>
>This is the reason why underwater communication is hard.
>The channel characteristics can change substantially over
>a small area, and over a short time.
>
>Within a couple of meters, you can loose or gain yet another
>multipath. Within minutes, the tide can change the depth
>enough to signifcicantly change the multipath pattern.
>
>>=A0That's
>> a nearly impossible task unless you're willing to spend a LOT of power
>> and bandwidth overhead on pilot tones and/or other mechanisms to
>> facilitate faster adaptation.
>
>Under water, you con't have much bandwidth. First of all,
>absorption losses are proportional to frequency. Scattering
>also becomes worse at high frequencies, so you want to
>work in low frequency bands. But then geometric effects
>come in to play, with normal mode interference from
>both sea surface and the bottom.
>
>And of course, the andwidth you do have, is contamicated
>by more or less random interference caused by severe
>multipath propagation.
>
>Rune
>

Thanks Rune

I am glad somebody understands what I am doing... :-)

Chintan