> Randy Yates <yates@ieee.org> wrote:
>
>> And, as in other types of receivers, "correlation detector"
>> and "matched filter" are synonymous given that the output is
>> sampled at the proper time.
>
> I'm sure you're right, but I usually think of a matched filter
> as being a convolution with a pulse shape, selected
> to correspond the transmitted pulse shape in a way that
> gives a maximum-likelyhood-ratio estimate. "Correlation
> method" to me is a broader term, involving correlation
> with anything, including a continuous-time quadrature signal
> (not a pulse).
> I could be mistaken; or in different settings the jargon
> is used inconsistently.
Point taken, Steve. I should know by now to avoid semantics arguments;
who cares as long as we know what is meant?! And as you say, usage
certainly varies.
When would you use a "correlation method" in which the correlation
signal isn't the desired signal?
--
% Randy Yates % "Rollin' and riding and slippin' and
%% Fuquay-Varina, NC % sliding, it's magic."
%%% 919-577-9882 %
%%%% <yates@ieee.org> % 'Living' Thing', *A New World Record*, ELO
http://www.digitalsignallabs.com
Reply by Jerry Avins●January 11, 20092009-01-11
Rafael Deliano wrote:
... slicer ( "Entscheider" )....
Perhaps less colorful, but more straightforward. It means "decider".
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by Steve Pope●January 11, 20092009-01-11
Randy Yates <yates@ieee.org> wrote:
> And, as in other types of receivers, "correlation detector"
> and "matched filter" are synonymous given that the output is
> sampled at the proper time.
I'm sure you're right, but I usually think of a matched filter
as being a convolution with a pulse shape, selected
to correspond the transmitted pulse shape in a way that
gives a maximum-likelyhood-ratio estimate. "Correlation
method" to me is a broader term, involving correlation
with anything, including a continuous-time quadrature signal
(not a pulse).
I could be mistaken; or in different settings the jargon
is used inconsistently.
Steve
Reply by Randy Yates●January 11, 20092009-01-11
"John E. Hadstate" <jh113355@hotmail.com> writes:
> "Tomeu" <tomeu@marexi.com> wrote in message
> news:u8Wdne20DZ9flPrUnZ2dnUVZ_sTinZ2d@giganews.com...
>> >
>>>
>>>> Hello all,
>>>>
>>>> The modulation scheme I am using is a non-coherent FSK.
>>>> The carrier frequencies are 20kHz and 22kHz.
>>>>
>>>> bandpass FIR filters at 20 and 22kHz respectively.
>>>>
>>>> The question is, which should be a typical number
>>>> of coefficients for those filters?
>>
>> Any suggestions are greatly appreciated.
>>
>
> I think the technique you are describing is called a "correlation
> detector for FSK."
First of all, I don't think it is. I think the OP was simply using a
pair of bandpass filters and hoping to get better performance by
narrowing the bandwidth.
And, as in other types of receivers, "correlation detector" and "matched
filter" are synonymous given that the output is sampled at the proper
time.
> How well it works compared to other methods depends on the symbol rate
> compared to the frequency deviation.
No. This method will give you the *best* BER performance (in AWGN) of
any method (or at least no worse).
> You can alleviate some of the bandpass filtering problems by
> converting your incoming stream to complex, splitting this into two
> streams (one for mark, the other for space), shifting the mark and
> space streams to 0 Hz, and then low pass filtering each of them. The
> beauty of this is that you can use shorter FIR filters to do the
> filtering, and that you can use the same coefficients in each filter.
I can see how that can possibly reduce complexity, but it won't yield
a better receiver in terms of BER.
The optimum receiver will use a filter length equal to the symbol
period. If you narrow the filters more (extend the FIR length), you
reduce the peak of the detected energy.
> Also, take a look at the following paper.
http://www.stw.nl/NR/rdonlyres/E7D3FA69-5103-4ADC-9D76-E575ED6DEC5C/0/lopelli.pdf
NICE paper! I especially like Figure 5 which shows how frequency offset
affects each method.
--Randy
PS: The optimum receiver I'm referring to is shown in Figure 2 of that
paper (albeit in a baseband (quadrature) receiver architecture rather
than bandpass).
--
% Randy Yates % "Remember the good old 1980's, when
%% Fuquay-Varina, NC % things were so uncomplicated?"
%%% 919-577-9882 % 'Ticket To The Moon'
%%%% <yates@ieee.org> % *Time*, Electric Light Orchestra
http://www.digitalsignallabs.com
Reply by Rafael Deliano●January 11, 20092009-01-11
> There is also an autocorrelation method for FSK.
Popular for modems on voicegrade lines. Less
appropriate for noisy radio/sonar.
The version with the two bandpass filters was used
by Siemens on a DSP in an early NMT mobile radio in 1984
( text in german, pictures should be obvious ):
http://www.embeddedFORTH.de/temp/bpfsk.pdf
The text notes that all the filters were optimized
to get good eyepatterns for the slicer ( "Entscheider" ).
The filters were probably implemented as wave digital
filters.
I think this version is a left over from the
analog days. Even if his carrier was in audio range
and not 21kHz.
MfG JRD
Reply by Steve Pope●January 11, 20092009-01-11
John E. Hadstate <jh113355@hotmail.com> wrote:
>"Tomeu" <tomeu@marexi.com> wrote in message
>>>> The modulation scheme I am using is a non-coherent FSK.
>>>> The carrier frequencies are 20kHz and 22kHz.
>>>> bandpass FIR filters at 20 and 22kHz respectively.
>I think the technique you are describing is called a
>"correlation detector for FSK."
I use different terminology. The correlation method is
used to detect coherent FSK, and treats the carrier frequencies
as orthogonal to each other and to all noise components,
and correlates each carrier to a quadrature signal. (Along
the lines of a typical demodulator for OFDM, but ony a couple of
carriers are populated.) This method starts to lose its
appeal once the FSK is decoherent.
Of course I could be completely mistaken as to the common
terminology.
There is also an autocorrelation method for FSK. There
was a discussion here on the topic, roughly four months
ago.
>How well it works compared to
>other methods depends on the symbol rate compared to the
>frequency deviation. You can alleviate some of the bandpass
>filtering problems by converting your incoming stream to
>complex, splitting this into two streams (one for mark, the
>other for space), shifting the mark and space streams to 0 Hz,
>and then low pass filtering each of them. The beauty of this
>is that you can use shorter FIR filters to do the filtering,
>and that you can use the same coefficients in each filter.
Reply by John E. Hadstate●January 11, 20092009-01-11
"Tomeu" <tomeu@marexi.com> wrote in message
news:u8Wdne20DZ9flPrUnZ2dnUVZ_sTinZ2d@giganews.com...
> >
>>
>>> Hello all,
>>>
>>> The modulation scheme I am using is a non-coherent FSK.
>>> The carrier frequencies are 20kHz and 22kHz.
>>>
>>> bandpass FIR filters at 20 and 22kHz respectively.
>>>
>>> The question is, which should be a typical number
>>> of coefficients for those filters?
>
> Any suggestions are greatly appreciated.
>
I think the technique you are describing is called a
"correlation detector for FSK." How well it works compared to
other methods depends on the symbol rate compared to the
frequency deviation. You can alleviate some of the bandpass
filtering problems by converting your incoming stream to
complex, splitting this into two streams (one for mark, the
other for space), shifting the mark and space streams to 0 Hz,
and then low pass filtering each of them. The beauty of this
is that you can use shorter FIR filters to do the filtering,
and that you can use the same coefficients in each filter.
Also, take a look at the following paper.
http://www.stw.nl/NR/rdonlyres/E7D3FA69-5103-4ADC-9D76-E575ED6DEC5C/0/lopelli.pdf
Reply by Randy Yates●January 9, 20092009-01-09
"Tomeu" <tomeu@marexi.com> writes:
> Hello all,
>
> I am involved in the development of an underwater modem. Right now I am
> dealing with the simulation stage with simulink.
>
> The modulation scheme I am using is a non-coherent FSK. The carrier
> frequencies are 20kHz and 22kHz.
>
> At the demodulator part, I have designed a matched filters scheme, with
> both bandpass FIR filters at 20 and 22kHz respectively. The question is,
> which should be a typical number of coefficients for those filters? I am
> not sure whether I make the filters too selective...
Hi Tomeu,
I think you meant here you're using just a couple of bandpass filters,
not "matched filters," right?
[leeandmesserschmitt3] derive the optimum receiver for non-coherent FSK
(see Figure 7-16). You use four matched filters, two for each
frequency, and then take the magnitude squared of each (you have to have
timing so you know when to sample).
The filter's coefficients are sin/cos functions at the desired
frequencies with an envelope of the pulse shape. If you're just using
rectangular pulse shaping, then the filters are simply sin/cos functions
with a length of the baud period.
Hope this helps.
--Randy
@BOOK{leeandmesserschmitt3,
title = "{Digital Communication}",
author = "{John R. Barry and Edwared A. Lee and David G. Messerschmitt}",
publisher = "Springer",
edition = "third",
year = "2004"}
>
> Otherwise, for non coherent detection, which scheme should be more robust,
> matched filters or a quadrature receiver?
>
> Thanks in advance
--
% Randy Yates % "With time with what you've learned,
%% Fuquay-Varina, NC % they'll kiss the ground you walk
%%% 919-577-9882 % upon."
%%%% <yates@ieee.org> % '21st Century Man', *Time*, ELO
http://www.digitalsignallabs.com
Reply by Vladimir Vassilevsky●January 9, 20092009-01-09
Tomeu wrote:
> Anyways. I had a misconception on matched filters. What I am doing right
> now is: two bandpass filters, one at 20 and one at 22kHz, then envelope
> detection. My question was: are this type of filters feasible? they have to
> be too selective.
Those filters are definitely not for the spoon-fed students. They should
use one filter from 20 to 22kHz, with a frequency counter after that.
> Any suggestions are greatly appreciated.
How much is the "greatly appreciation"?
VLV
Reply by ●January 9, 20092009-01-09
On Jan 9, 5:23�am, spop...@speedymail.org (Steve Pope) wrote:
> Tomeu <to...@marexi.com> wrote:
> >Anyways. I had a misconception on matched filters. What I am doing right
> >now is: two bandpass filters, one at 20 and one at 22kHz, then envelope
> >detection. My question was: are this type of filters feasible? they have to
> >be too selective.
>
> That can work if the bauds are long enough. �I'm not sure
> you have said how fast you are signalling, or maybe I missed it.
>
> Steve
X-no-archive:
You may want to research the old ham radio system called RTTY. For a
low baud rate system like that, it was seen that using two separate
filters for mark and space can work better becasue it rejects the
interference in between the two tones. That only works if the
frequency spacing is high relative to the baud rate. I could not find
much on the web about this other than here:
http://lists.contesting.com/_rtty/2001-09/msg00027.html
Mark