FSK Versus OOK Demodulation

On Jun 21, 6:34=A0pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:

>
> Doctor Dilip Matlabi,
>
> Forget about minor technical details of crippled algorithms for a
> moment. Optimal is about maximum likelihood and conditional
> probabilities. The only difference between optimal (2 x OOK) and optimal
> FSK is that the phase relationship between "mark" and "space" is not set
> with OOK.

Bullshit!

On 06/21/2011 05:37 PM, dvsarwate wrote:
> On Jun 21, 2:51 pm, Randy Yates<ya...@ieee.org>  asked:
>> With the right filtering, an FSK signal can be viewed as two
>> complementary OOK (on-off keyed) signals. Is the optimal FSK
>> demodulator more optimal, less optimal, or equivalent to two optimal
>> OOK demodulators with their outputs combined?
>
> An optimal FSK demodulator compares the
> outputs of two filters F0 and F1 and decides
> 0 or 1 depending on which filter has the larger
> output.  An optimal OOK demodulator compares
> the output of a filter to a fixed threshold and
> decides ON or OFF depending on whether the
> output exceeds the threshold or not.  And while
> we are at it, the optimum choice of threshold is
> NOT the half the average output when the ON
> signal is being transmitted.  So now consider
> two optimal OOK demodulators with filters F0
> and F1 respectively  (these filters are the same
> as the ones the optimal FSK demodulator is
> using).  Suppose the (optimal) threshold is 0.4.
> Suppose also that a 0 is being transmitted.
> If filter F0 has output 1.3 while filter F1 has
> output 0.45, then (since 1.3>  0.45), the
> optimal FSK demodulator correctly decides
> that 0 is being transmitted.  But the two
> optimal OOK detectors both think that their
> respective signals are ON.  How do we now
> combine these ON outputs to determine whether
> a 0 or a 1 was transmitted?
>
> Oh you mean we look at the outputs 1.3 and 0.4
> of the two filters and choose the larger?  and
> where are the *optimal OOK detectors* when you
> do so?  What are their thresholds? etc.
>
> Making decisions *individually* from the filter
> outputs (and thereby ignoring the complementary
> aspect of the signals) as is done by the optimal
> OOK detector does not work for demodulating
> FSK.  If a joint decision is being made, then the
> optimal OOK detector has disappeared, and the
> question asked is moot.
>
> Hope this helps.

Dilip,

Thank you for replying. It makes perfect sense, but there's a little
voice in my head asking, "How do we know there is NOT some other
method of processing the two OOK outputs that yields something just as
useful?"
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO

On 06/21/2011 05:37 PM, dvsarwate wrote:
> [...]
> Making decisions *individually* from the filter outputs (and thereby
> ignoring the complementary aspect of the signals) as is done by the
> optimal OOK detector does not work for demodulating FSK.

Doesn't work? It would work, just not optimally - correct? That is, if
you just selected one of the two tones and decoded it as OOK, it would
certainly "work," would it not?

Actually, let me tell you why this whole OOK vs FSK thing is an
important question to me. In the actual system I'm dealing with,
multipath with a large delay spread can occur causing
frequency-selective fading of one of the FSK tones.

In that case there are at least two ways to proceed: 1) decode the
stronger of the two as OOK and just drop the other, or 2) decode both
as OOK and somehow "optimally" combine them.

First principles of information theory tells us it's going to be
better to utilize both OOKs (except in the extreme case in which the
signal is completely gone) because there's more, well, information
there.

Comments?
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO

On 06/21/2011 09:20 PM, Randy Yates wrote:
> [...]
> 2) decode both as OOK and somehow "optimally" combine them.

PS: Assume you have a weighting vector w, -1 <= w <= +1, that
indicates the relative "goodness" of each OOK, 0 being equally
good.
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO

On 06/21/2011 06:41 PM, Tim Wescott wrote:
> [...]
> Even without phase continuity it's more optimal in that it doesn't
> make the receiver guess at a zero level

Tim, what "zero level" are you referring to? Or did you mean "guess at a threshold?"

> I suspect that any FSK you're going to run across today will be of
> continuous phase, but I'm not going to _trust_ that unless I see it
> in the specification!

Yes, I neglected to specify that we should consider this discontinuous-phase
FSK.
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO

On 06/21/2011 07:27 PM, Randy Yates wrote:
> On 06/21/2011 06:41 PM, Tim Wescott wrote:
>> [...]
>> Even without phase continuity it's more optimal in that it doesn't
>> make the receiver guess at a zero level
>
> Tim, what "zero level" are you referring to? Or did you mean "guess at a
> threshold?"

Guess at a threshold -- actually, for decent OOK reception, the receiver 
needs to make a guess at the pulse amplitude, so it can slice it 
effectively.

>> I suspect that any FSK you're going to run across today will be of
>> continuous phase, but I'm not going to _trust_ that unless I see it
>> in the specification!
>
> Yes, I neglected to specify that we should consider this
> discontinuous-phase
> FSK.


-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

Randy Yates wrote:

> On 06/21/2011 09:20 PM, Randy Yates wrote:
> 
>> [...]
>> 2) decode both as OOK and somehow "optimally" combine them.
> 
> 
> PS: Assume you have a weighting vector w, -1 <= w <= +1, that
> indicates the relative "goodness" of each OOK, 0 being equally
> good.

Then the constellation points are (w+1,0) and (0, w-1), on assumption 
that the FSK is orthogonal. Now connect the points with a piece of 
straight line and build a middle perpendicular to that piece. This 
perpendicular is the decision border between "1" and "0".


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

>On 06/21/2011 02:13 PM, Vladimir Vassilevsky wrote:
>>
>>
>> Randy Yates wrote:
>>
>>> With the right filtering, an FSK signal can be viewed as two
>>> complementary OOK (on-off keyed) signals.
>>
>> There is more than that. The phase is continuous.
>
>Not necessarily, unless phase continuity is specified (and the 
>specification complied to).  Old teletype FSK was sometimes accomplished 
>with a pair of audio-frequency oscillators, their outputs switched, and 
>fed to a SSB transmitter.
>
>>> Is the optimal FSK
>>> demodulator more optimal, less optimal, or equivalent to two optimal
>>> OOK demodulators with their outputs combined?
>>
>> Optimal FSK demodulator is more optimal then two OOK demodulators as it
>> makes use of the continuity of the phase.
>
>Even without phase continuity it's more optimal in that it doesn't make 
>the receiver guess at a zero level -- for the case that I remember, a 
>sufficient statistic is the difference in signal strengths out of a pair 
>of matched filters.  In as much as the two signals are nominally of the 
>same amplitude, the sufficient statistic was a signed number with the 
>decision point at 0.
>
>I suspect that any FSK you're going to run across today will be of 
>continuous phase, but I'm not going to _trust_ that unless I see it in 
>the specification!

Many modern implementations of the PSTN FSK modems still do not ensure
continuous phase, so your lack of trust is fully justified.

Steve

I think we first have to define what we mean by "optimal"

Are we talking about an AWGN channel only? or are we talking about
more interesting channels?


The old ham radio RTTY systems used 2 OOK receivers combined and they
worked well under conditions like:

1) selective fading where F1 or F2 might dissappear

2) co-channel interference where there might be a strong CW tone
between F1 and F2. (depending on the baud rate and F1 to F2 spacing,
there can be a decent sized "gap" between F1 and F2.  Two narrow OOK
receivers can ignore what is in the gap while an FM demod cannot.)


It is interesting for me to see this discussion here, I can remember
vividly being a kid about 13 years old first exposed to RTTY systems
and looking at the schematic and wondering what makes the "optimal"
RTTY receiver... why they used 2 AM detecotrs instead of one singel FM
detector...  I'm glad to see it was not a simple question and I look
forward to reading what answer the group develops...

Mark

On 06/22/2011 08:32 AM, Mark wrote:
> I think we first have to define what we mean by "optimal"
>
> Are we talking about an AWGN channel only? or are we talking about
> more interesting channels?
>
>
> The old ham radio RTTY systems used 2 OOK receivers combined and they
> worked well under conditions like:
>
> 1) selective fading where F1 or F2 might dissappear
>
> 2) co-channel interference where there might be a strong CW tone
> between F1 and F2. (depending on the baud rate and F1 to F2 spacing,
> there can be a decent sized "gap" between F1 and F2.  Two narrow OOK
> receivers can ignore what is in the gap while an FM demod cannot.)
>
>
> It is interesting for me to see this discussion here, I can remember
> vividly being a kid about 13 years old first exposed to RTTY systems
> and looking at the schematic and wondering what makes the "optimal"
> RTTY receiver... why they used 2 AM detecotrs instead of one singel FM
> detector...  I'm glad to see it was not a simple question and I look
> forward to reading what answer the group develops...

Mark,

Indeed, this is a derivative of the old RTTY systems. It is the
GMDSS/DSC system. It is good to know I'm on the right track - I
had no idea this had been done before!
-- 
Randy Yates                      % "Watching all the days go by...
Digital Signal Labs              %  Who are you and who am I?"
mailto://yates@ieee.org          % 'Mission (A World Record)',
http://www.digitalsignallabs.com % *A New World Record*, ELO