Forums

Equalization for non-coherent FSK burst modem

Started by mite_learner January 27, 2014
I want to work on the equalization techniques that are best suited for
burst 2-4FSK modems using non-coherent detection. Data rate targeted is
0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but
couldn't find any thing on burst FSK modems. Can somebody please suggest me
some literature that target this. Also, if some body can point out any
existing implementation that works well so that I could implement right
thing, this would be very helpful. 


Best,
ML	 

_____________________________		
Posted through www.DSPRelated.com
"mite_learner" <94814@dsprelated> writes:

> I want to work on the equalization techniques that are best suited for > burst 2-4FSK modems using non-coherent detection. Data rate targeted is > 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but > couldn't find any thing on burst FSK modems. Can somebody please suggest me > some literature that target this. Also, if some body can point out any > existing implementation that works well so that I could implement right > thing, this would be very helpful.
Hi, You might check out "GSM Channel Equalization Algorithm - Modern DSP Coprocessor Approach," Milos Drutarovsky, Radioengineering, Vol. 8, No. 4, December 1999. I am not sure if it works for generic FSK (as opposed to MSK). He uses the approximation from Laurent's famous paper, which I would think generalizes to FSK, but I'm still reading... -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote:
> I want to work on the equalization techniques that are best suited for > > burst 2-4FSK modems using non-coherent detection. Data rate targeted is > > 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but > > couldn't find any thing on burst FSK modems. Can somebody please suggest me > > some literature that target this. Also, if some body can point out any > > existing implementation that works well so that I could implement right > > thing, this would be very helpful. > > > > > > Best, > > ML > > > > _____________________________ > > Posted through www.DSPRelated.com
One of the simplest equalization techniques for FSK is called the "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR filter which restores the constant-envelope property of the FSK signal. Check it out. John
John <sampson164@gmail.com> writes:

> On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: >> I want to work on the equalization techniques that are best suited for >> >> burst 2-4FSK modems using non-coherent detection. Data rate targeted is >> >> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but >> >> couldn't find any thing on burst FSK modems. Can somebody please suggest me >> >> some literature that target this. Also, if some body can point out any >> >> existing implementation that works well so that I could implement right >> >> thing, this would be very helpful. >> >> >> >> >> >> Best, >> >> ML >> >> >> >> _____________________________ >> >> Posted through www.DSPRelated.com > > One of the simplest equalization techniques for FSK is called the > "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR > filter which restores the constant-envelope property of the FSK > signal. Check it out.
John, I've heard that is a good technique too, but I've never used it. One potential problem in applying this technique to burst data is that the burst may not last long enough for the adaptation to converge. Have you heard of some sort of initialization (perhaps based on sync symbols) being incorporated into CMA? -- Randy Yates Digital Signal Labs http://www.digitalsignallabs.com
On Monday, January 27, 2014 7:55:07 AM UTC-5, Randy Yates wrote:
> John <sampson164@gmail.com> writes: > > > > > On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: > > >> I want to work on the equalization techniques that are best suited for > > >> > > >> burst 2-4FSK modems using non-coherent detection. Data rate targeted is > > >> > > >> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but > > >> > > >> couldn't find any thing on burst FSK modems. Can somebody please suggest me > > >> > > >> some literature that target this. Also, if some body can point out any > > >> > > >> existing implementation that works well so that I could implement right > > >> > > >> thing, this would be very helpful. > > >> > > >> > > >> > > >> > > >> > > >> Best, > > >> > > >> ML > > >> > > >> > > >> > > >> _____________________________ > > >> > > >> Posted through www.DSPRelated.com > > > > > > One of the simplest equalization techniques for FSK is called the > > > "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR > > > filter which restores the constant-envelope property of the FSK > > > signal. Check it out. > > > > John, > > > > I've heard that is a good technique too, but I've never used it. > > > > One potential problem in applying this technique to burst data is that > > the burst may not last long enough for the adaptation to converge. Have > > you heard of some sort of initialization (perhaps based on sync symbols) > > being incorporated into CMA? > > -- > > Randy Yates > > Digital Signal Labs > > http://www.digitalsignallabs.com
I haven't tried that but I'll bet it has been tried. If you can capture the entire burst then you can actually compute the taps once using matrix math instead of gradient descent.
On Monday, January 27, 2014 10:22:54 AM UTC-5, John wrote:
> On Monday, January 27, 2014 7:55:07 AM UTC-5, Randy Yates wrote: > > > John <sampson164@gmail.com> writes: > > > > > > > > > > > > > On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: > > > > > > >> I want to work on the equalization techniques that are best suited for > > > > > > >> > > > > > > >> burst 2-4FSK modems using non-coherent detection. Data rate targeted is > > > > > > >> > > > > > > >> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but > > > > > > >> > > > > > > >> couldn't find any thing on burst FSK modems. Can somebody please suggest me > > > > > > >> > > > > > > >> some literature that target this. Also, if some body can point out any > > > > > > >> > > > > > > >> existing implementation that works well so that I could implement right > > > > > > >> > > > > > > >> thing, this would be very helpful. > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> Best, > > > > > > >> > > > > > > >> ML > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> _____________________________ > > > > > > >> > > > > > > >> Posted through www.DSPRelated.com > > > > > > > > > > > > > > One of the simplest equalization techniques for FSK is called the > > > > > > > "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR > > > > > > > filter which restores the constant-envelope property of the FSK > > > > > > > signal. Check it out. > > > > > > > > > > > > John, > > > > > > > > > > > > I've heard that is a good technique too, but I've never used it. > > > > > > > > > > > > One potential problem in applying this technique to burst data is that > > > > > > the burst may not last long enough for the adaptation to converge. Have > > > > > > you heard of some sort of initialization (perhaps based on sync symbols) > > > > > > being incorporated into CMA? > > > > > > -- > > > > > > Randy Yates > > > > > > Digital Signal Labs > > > > > > http://www.digitalsignallabs.com > > > > I haven't tried that but I'll bet it has been tried. If you can capture the entire burst then you can actually compute the taps once using matrix math instead of gradient descent.
One common misconception of all these "algorithms" is that the objective function and solution are often thought to be one and the same. This is not the case. The CMA, just like many MMSE algorithms, can be solved simultaneously without using an adaptive filter. In fact when your observation length is very limited, it is much better to do "block-processing" than symbol-by-symbol adaptation.
>John <sampson164@gmail.com> writes: > >> On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: >>> I want to work on the equalization techniques that are best suited for >>> >>> burst 2-4FSK modems using non-coherent detection. Data rate targeted
is
>>> >>> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but >>> >>> couldn't find any thing on burst FSK modems. Can somebody please
suggest me
>>> >>> some literature that target this. Also, if some body can point out any >>> >>> existing implementation that works well so that I could implement
right
>>> >>> thing, this would be very helpful. >>> >>> >>> >>> >>> >>> Best, >>> >>> ML >>> >>> >>> >>> _____________________________ >>> >>> Posted through www.DSPRelated.com >> >> One of the simplest equalization techniques for FSK is called the >> "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR >> filter which restores the constant-envelope property of the FSK >> signal. Check it out. > >John, > >I've heard that is a good technique too, but I've never used it. > >One potential problem in applying this technique to burst data is that >the burst may not last long enough for the adaptation to converge. Have >you heard of some sort of initialization (perhaps based on sync symbols) >being incorporated into CMA? >-- >Randy Yates >Digital Signal Labs >http://www.digitalsignallabs.com >
Hi john, Thanks. I am going to deal with very short burst, maybe 20 to 30 symbols only, do you think CMA is going to work for such short burst. Any relevant material would be appreciated. ML _____________________________ Posted through www.DSPRelated.com
On Thursday, January 30, 2014 10:50:49 PM UTC-5, mite_learner wrote:
> >John <sampson164@gmail.com> writes: > > > > > >> On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: > > >>> I want to work on the equalization techniques that are best suited for > > >>> > > >>> burst 2-4FSK modems using non-coherent detection. Data rate targeted > > is > > >>> > > >>> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc but > > >>> > > >>> couldn't find any thing on burst FSK modems. Can somebody please > > suggest me > > >>> > > >>> some literature that target this. Also, if some body can point out any > > >>> > > >>> existing implementation that works well so that I could implement > > right > > >>> > > >>> thing, this would be very helpful. > > >>> > > >>> > > >>> > > >>> > > >>> > > >>> Best, > > >>> > > >>> ML > > >>> > > >>> > > >>> > > >>> _____________________________ > > >>> > > >>> Posted through www.DSPRelated.com > > >> > > >> One of the simplest equalization techniques for FSK is called the > > >> "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR > > >> filter which restores the constant-envelope property of the FSK > > >> signal. Check it out. > > > > > >John, > > > > > >I've heard that is a good technique too, but I've never used it. > > > > > >One potential problem in applying this technique to burst data is that > > >the burst may not last long enough for the adaptation to converge. Have > > >you heard of some sort of initialization (perhaps based on sync symbols) > > >being incorporated into CMA? > > >-- > > >Randy Yates > > >Digital Signal Labs > > >http://www.digitalsignallabs.com > > > > > > > Hi john, > > > > Thanks. I am going to deal with very short burst, maybe 20 to 30 symbols > > only, do you think CMA is going to work for such short burst. Any relevant > > material would be appreciated. > > > > ML > > > > _____________________________ > > Posted through www.DSPRelated.com
For the block version you can use something like this in Matlab. What we're doing here is basically solving the W-H Equation for the filter taps. function zeq = cma(z,Neq) [dummy,R]=corrmtx(z,Neq-1); P=xcorr(z,z./abs(z),(Neq-1)/2); W=R\P;W=W/sum(abs(W)); zeq = filter(W,1,z); end John
>On Thursday, January 30, 2014 10:50:49 PM UTC-5, mite_learner wrote: >> >John <sampson164@gmail.com> writes: >> >> > >> >> >> On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: >> >> >>> I want to work on the equalization techniques that are best suited
for
>> >> >>> >> >> >>> burst 2-4FSK modems using non-coherent detection. Data rate
targeted
>> >> is >> >> >>> >> >> >>> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc
but
>> >> >>> >> >> >>> couldn't find any thing on burst FSK modems. Can somebody please >> >> suggest me >> >> >>> >> >> >>> some literature that target this. Also, if some body can point out
any
>> >> >>> >> >> >>> existing implementation that works well so that I could implement >> >> right >> >> >>> >> >> >>> thing, this would be very helpful. >> >> >>> >> >> >>> >> >> >>> >> >> >>> >> >> >>> >> >> >>> Best, >> >> >>> >> >> >>> ML >> >> >>> >> >> >>> >> >> >>> >> >> >>> _____________________________ >> >> >>> >> >> >>> Posted through www.DSPRelated.com >> >> >> >> >> >> One of the simplest equalization techniques for FSK is called the >> >> >> "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR >> >> >> filter which restores the constant-envelope property of the FSK >> >> >> signal. Check it out. >> >> > >> >> >John, >> >> > >> >> >I've heard that is a good technique too, but I've never used it. >> >> > >> >> >One potential problem in applying this technique to burst data is that >> >> >the burst may not last long enough for the adaptation to converge.
Have
>> >> >you heard of some sort of initialization (perhaps based on sync
symbols)
>> >> >being incorporated into CMA? >> >> >-- >> >> >Randy Yates >> >> >Digital Signal Labs >> >> >http://www.digitalsignallabs.com >> >> > >> >> >> >> Hi john, >> >> >> >> Thanks. I am going to deal with very short burst, maybe 20 to 30
symbols
>> >> only, do you think CMA is going to work for such short burst. Any
relevant
>> >> material would be appreciated. >> >> >> >> ML >> >> >> >> _____________________________ >> >> Posted through www.DSPRelated.com > >For the block version you can use something like this in Matlab. What
we're doing here is basically solving the W-H Equation for the filter taps.
> >function zeq = cma(z,Neq) > > [dummy,R]=corrmtx(z,Neq-1); > P=xcorr(z,z./abs(z),(Neq-1)/2); > W=R\P;W=W/sum(abs(W)); > zeq = filter(W,1,z); > >end > >John >
Thanks for you reply. I am also planning to do frequency hopping. Would that benefit me against multipath distortion? ML _____________________________ Posted through www.DSPRelated.com
On Thu, 06 Feb 2014 10:25:24 -0600, "mite_learner" <94814@dsprelated>
wrote:

>>On Thursday, January 30, 2014 10:50:49 PM UTC-5, mite_learner wrote: >>> >John <sampson164@gmail.com> writes: >>> >>> > >>> >>> >> On Monday, January 27, 2014 2:50:11 AM UTC-5, mite_learner wrote: >>> >>> >>> I want to work on the equalization techniques that are best suited >for >>> >>> >>> >>> >>> >>> burst 2-4FSK modems using non-coherent detection. Data rate >targeted >>> >>> is >>> >>> >>> >>> >>> >>> 0.5Mbps to 1 Mbps. I found many stuff on equalization for QPSK etc >but >>> >>> >>> >>> >>> >>> couldn't find any thing on burst FSK modems. Can somebody please >>> >>> suggest me >>> >>> >>> >>> >>> >>> some literature that target this. Also, if some body can point out >any >>> >>> >>> >>> >>> >>> existing implementation that works well so that I could implement >>> >>> right >>> >>> >>> >>> >>> >>> thing, this would be very helpful. >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> Best, >>> >>> >>> >>> >>> >>> ML >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> >>> _____________________________ >>> >>> >>> >>> >>> >>> Posted through www.DSPRelated.com >>> >>> >> >>> >>> >> One of the simplest equalization techniques for FSK is called the >>> >>> >> "Constant Modulus Algorithm" or CMA. CMA adaptively derives a FIR >>> >>> >> filter which restores the constant-envelope property of the FSK >>> >>> >> signal. Check it out. >>> >>> > >>> >>> >John, >>> >>> > >>> >>> >I've heard that is a good technique too, but I've never used it. >>> >>> > >>> >>> >One potential problem in applying this technique to burst data is that >>> >>> >the burst may not last long enough for the adaptation to converge. >Have >>> >>> >you heard of some sort of initialization (perhaps based on sync >symbols) >>> >>> >being incorporated into CMA? >>> >>> >-- >>> >>> >Randy Yates >>> >>> >Digital Signal Labs >>> >>> >http://www.digitalsignallabs.com >>> >>> > >>> >>> >>> >>> Hi john, >>> >>> >>> >>> Thanks. I am going to deal with very short burst, maybe 20 to 30 >symbols >>> >>> only, do you think CMA is going to work for such short burst. Any >relevant >>> >>> material would be appreciated. >>> >>> >>> >>> ML >>> >>> >>> >>> _____________________________ >>> >>> Posted through www.DSPRelated.com >> >>For the block version you can use something like this in Matlab. What >we're doing here is basically solving the W-H Equation for the filter >taps. >> >>function zeq = cma(z,Neq) >> >> [dummy,R]=corrmtx(z,Neq-1); >> P=xcorr(z,z./abs(z),(Neq-1)/2); >> W=R\P;W=W/sum(abs(W)); >> zeq = filter(W,1,z); >> >>end >> >>John >> > >Thanks for you reply. > >I am also planning to do frequency hopping. Would that benefit me against >multipath distortion? > >ML
Not much. It really only helps if you would have otherwise picked a channel that was in a deep fade. If the system is smart enough to not do that, and/or to move channels when it loses lock due to a fade, then FH isn't going to help to mitigate frequency-selective fading. If the signal bandwidth is wider than the channel coherence bandwidth, it'll help even less. FH is mostly an interference mitigation/suppression technique. Eric Jacobsen Anchor Hill Communications http://www.anchorhill.com