Frequency offset compensation for 802.15.4 (ZigBee-MSK)

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>On Tue, 6 Oct 2015 20:14:18 +0000 (UTC), spope33@speedymail.org (Steve

>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>>> On Tue, 6 Oct 2015 03:34:48 +0000 (UTC), spope33@speedymail.org (Steve

>>>> Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>>>>> If you apply any kind of filtering to MSK, it's not MSK any more.

>>>> Really it comes down to your being willing to call distorted
>>>> PSK "PSK", but unwilling to call distorted MSK "MSK".

>>> What "distortion" of MSK are you talking about?

>>Exactly what you described above.

>Are you disagreeing that adding a filter that changes an MSK waveform
>makes it deviate from being MSK?   

Not at all.  It no longer meets the exact definition of MSK,
once filtered.

> If it can change, how much can it change before it's not MSK?

That's a terminology question, not a question about the facts.

>>> you've been stating some things as facts that simply aren't.
>>
>>If I have stated anything non-factual, I will correct it, but
>>right now I don't believe there are any instances of such.

>You seem to be claiming that filtering PSK makes it not PSK.   

That is not exactly what I said.  I did say if one is choosing to
say that filtering MSK can make it "not MSK", one can equally say that 
filtering PSK can make it "not PSK".

I then proceeded to describe that if the filtering of a PSK signal 
introduces amplitude modulation, the signal no longer meets the exact 
definition of being PSK, which would have phase modulation only.

You are now in a position entirely analogous to that of having
filtered MSK, and thereby not meeting the exact definition of MSK.

> Filters are integral parts of PSK systems to achieve matched 
> filtering performance.   How do you reconcile that?

I see nothing to reconcile.

>You also claimed that adding a filter to a phase modulated signal adds
>amplitude distortion.   

I said that it _can_ result in amplitude _modulation_ .

With unfiltered PSK (by that I mean, in which the pulse shape is an 
impulse), each symbol interval contains the same energy regardless
of data pattern.

Filters with impulse responses longer than one symbol can introduce 
A.M. in the sense that the symbol intervals will contain different 
amounts of energy, in a data-dependent fashion.  You now have a signal 
that is both amplitude- and phase-modulated, as I illustrated with
the example of applying 1 + D to QPSK.

In some cases, A.M. that exists on the transmitted signal can be
removed by processing in the receiver, but that is utterly irrelevant
to this discussion ... the type of modulation is determined solely by
the waveform of the transmitted signal, not by the receiver signal 
processing.

Hope this helps.

Steve

On Tue, 6 Oct 2015 22:54:47 +0000 (UTC), spope33@speedymail.org (Steve
Pope) wrote:

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>On Tue, 6 Oct 2015 20:14:18 +0000 (UTC), spope33@speedymail.org (Steve
>
>>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>>> On Tue, 6 Oct 2015 03:34:48 +0000 (UTC), spope33@speedymail.org (Steve
>
>>>>> Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>>>>> If you apply any kind of filtering to MSK, it's not MSK any more.
>
>>>>> Really it comes down to your being willing to call distorted
>>>>> PSK "PSK", but unwilling to call distorted MSK "MSK".
>
>>>> What "distortion" of MSK are you talking about?
>
>>>Exactly what you described above.
>
>>Are you disagreeing that adding a filter that changes an MSK waveform
>>makes it deviate from being MSK?   
>
>Not at all.  It no longer meets the exact definition of MSK,
>once filtered.

We're in agreement there.

>> If it can change, how much can it change before it's not MSK?
>
>That's a terminology question, not a question about the facts.
>
>>>> you've been stating some things as facts that simply aren't.
>>>
>>>If I have stated anything non-factual, I will correct it, but
>>>right now I don't believe there are any instances of such.
>
>>You seem to be claiming that filtering PSK makes it not PSK.   
>
>That is not exactly what I said.  I did say if one is choosing to
>say that filtering MSK can make it "not MSK", one can equally say that 
>filtering PSK can make it "not PSK".

What you said was:

>>Eric: If you apply any kind of filtering to MSK, it's not MSK any more.

>Steve: True.  The exact same could be said about PSK.   

The above is not a "can" or "may" statement.  This statement says that
any filtering changes PSK to be something other than PSK.   This
misses the fact that MSK and PSK are different in this regard.
Filtering PSK does not make it not PSK.   Filtering is necessary in
PSK to achieve matched filter performance.   

>I then proceeded to describe that if the filtering of a PSK signal 
>introduces amplitude modulation, the signal no longer meets the exact 
>definition of being PSK, which would have phase modulation only.

What you said was:

>I will counter by saying that, in the general case, filtering 
>a phase-modulated signal will add amplitude modulation, thereby moving it 
>outside of the definition of being purely phase modulated.

That's not "if" it adds AM, you said it "will" add AM, and will
therefore no longer be PSK.  But, even if it does add AM between the
symbols, it is still PSK.

>You are now in a position entirely analogous to that of having
>filtered MSK, and thereby not meeting the exact definition of MSK.

MSK and PSK differ significantly in this regard.

Most matched filter pulse shapes used in PSK do add some AM, but in
the receiver it is between the symbol sampling instances and therefore
adds no AM to the "modulation" or the sliced constellation.   The
ability to adjust those filter shapes for varying excess bandwidths
and still maintain matched filter performance is a big reason why PSK
is often used instead of MSK or other CPM variants, where the spectral
occupancy is much wider and a power efficiency penalty is paid
whenever you try to improve it.   PSK differs from CPM in that regard
and does not suffer from that problem, and so applying matched
filtering, which increases the intra-symbol AM and PAPR the more you
narrow the spectrum, does not add AM to the sliced constellation in
the receiver, does not reduce power efficiency, and does not make the
signal something other than PSK.

So, an infinite variety of filters can be selected which do not add AM
to the PSK modulation constellation in the receiver.   Also, even
introducing or increasing AM in the intra-symbol regions, which
generally happens even in a zero-ISI matched filter when the excess
bandwidth is reduced, does still meet the exact definition of being
PSK, which is still phase-only modulated at the slicer, i.e., the
phase carries the information.

This is contrary to what you claimed above.

>> Filters are integral parts of PSK systems to achieve matched 
>> filtering performance.   How do you reconcile that?
>
>I see nothing to reconcile.
>
>>You also claimed that adding a filter to a phase modulated signal adds
>>amplitude distortion.   
>
>I said that it _can_ result in amplitude _modulation_ .

Again, what you said was:

>I will counter by saying that, in the general case, filtering 
>a phase-modulated signal will add amplitude modulation, thereby moving it 
>outside of the definition of being purely phase modulated.

There's a "will" in there, not a "can".

But even if there is AM in the intra-symbol regions, which there often
is with PSK, it is still PSK.   CPM, including MSK, on the other hand,
is always constant envelope.   This is an important distinction that
makes your assertion incorrect.

MSK/CPM comes with limitations that PSK is not burdened with.
Spectral shaping via filtering without a power efficiency penalty is
one of them.   MSK and PSK are different in this regard.   It is not
possible to apply a filter, in the phase domain or in the IQ signal
stream, that changes the waveform and still call it MSK.   Because PSK
is generally NOT generated with a VCO, but with a quadrature signal
path, there is no phase filter and the IQ data paths can be filtered
aggressively without loss of performance.   Even if that aggressive
filter adds a lof of AM, if the filter in the receiver matches it (in
a matched filter sense), there will be no loss of performance or
addition of AM to the sliced constellation, and the signal will still
be every bit as much of a PSK system as if a different filter was
used.  

>With unfiltered PSK (by that I mean, in which the pulse shape is an 
>impulse), each symbol interval contains the same energy regardless
>of data pattern.

>Filters with impulse responses longer than one symbol can introduce 
>A.M. in the sense that the symbol intervals will contain different 
>amounts of energy, in a data-dependent fashion.  You now have a signal 
>that is both amplitude- and phase-modulated, as I illustrated with
>the example of applying 1 + D to QPSK.

That seems a bit to me like saying 12VDC systems won't run well on
120VAC.  Regardless, most effective bandlimiting pulse matched filters
traverse many symbols.   Even mild RC filters, like a 40% EBW filter
need around eight symbols in duration for the pulse.   Reducing the
spectral width to 10% EBW or so makes it even longer.  A mod/demod
pair with a matching set of those will see no AM or ISI in the sliced
constellation, and will still be a PSK system, even though it has been
filtered, and even though it may have AM in the intra-symbol regions.
The constellations will be the same.

This is why I asked the other question regarding this plot:

http://www.ericjacobsen.org/Files/8PSKconstellation.jpg

of an 8PSK constellation in a receiver.   You can't tell how it was
filtered, or how aggressively, because it won't change the
constellation as long as the filters match.   The amount of AM in the
signal changes with the selected EBW, and it has zero affect on the
received constellation.   That's still a PSK constellation, regardless
of how aggressive the filtering was, how many symbols the pulse
traversed, or how much AM there was between symbols.

>In some cases, A.M. that exists on the transmitted signal can be
>removed by processing in the receiver, but that is utterly irrelevant
>to this discussion ... the type of modulation is determined solely by
>the waveform of the transmitted signal, not by the receiver signal 
>processing.
>
>Hope this helps.
>
>Steve

Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>On Tue, 6 Oct 2015 22:54:47 +0000 (UTC), spope33@speedymail.org (Steve

>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>>>Are you disagreeing that adding a filter that changes an MSK waveform
>>>makes it deviate from being MSK?   

>>Not at all.  It no longer meets the exact definition of MSK,
>>once filtered.

>We're in agreement there.

>>>You seem to be claiming that filtering PSK makes it not PSK.   

>>That is not exactly what I said.  I did say if one is choosing to
>>say that filtering MSK can make it "not MSK", one can equally say that 
>>filtering PSK can make it "not PSK".

>What you said was:

>>>Eric: If you apply any kind of filtering to MSK, it's not MSK any more.

>>Steve: True.  The exact same could be said about PSK.   

>The above is not a "can" or "may" statement.  This statement says that
>any filtering changes PSK to be something other than PSK.   

I said it "could be said" that filtering changes PSK into something other 
than PSK.  Saying "is could be said X" is not the same as saying "X".

>This misses the fact that MSK and PSK are different in this regard.

Well, I agree they are different in this regard, but not as different
as you think.  

MSK, by definition, has a pulse shape that is non-zero for exactly one 
symbol.  PSK may or may not; but in those cases where it does
(which is very common), adding additional filtering such that the
overall filter response is longer than a symbol introduces A.M into
the signal.  From this perspective MSK and PSK are behaving similarly
under added filtering.

> Filtering is necessary in PSK to achieve matched filter performance.   

>What you said was:

>>I will counter by saying that, in the general case, filtering 
>>a phase-modulated signal will add amplitude modulation, thereby moving it 
>>outside of the definition of being purely phase modulated.

>That's not "if" it adds AM, you said it "will" add AM, and will
>therefore no longer be PSK.  

No, you are not paraphrasing me correctly, due to my having used
the qualifying language "in the general case".  

Consider:

"In the general case, real numbers are not rational." (true statement)
"real numbers are not rational"  (false statement)

>Most matched filter pulse shapes used in PSK do add some AM, but in
>the receiver it is between the symbol sampling instances and therefore
>adds no AM to the "modulation" or the sliced constellation.   

What counts in this discussion is the waveform of the signal being 
transmitted.  Processing within the receiver doesn't count.

>So, an infinite variety of filters can be selected which do not add AM
>to the PSK modulation constellation in the receiver.   

Sure.  But I have not been talking about signals deep within some
receiver algorithm.  

Suppose I construct an algorithm that takes a PAM signal and outputs
an FSK signal; shall I then be able to claim that, by transmitting the 
PAM signal, I was really transmitting FSK?

Steve

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>On Tue, 6 Oct 2015 22:54:47 +0000 (UTC), spope33@speedymail.org (Steve
>
>>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>>>Are you disagreeing that adding a filter that changes an MSK waveform
>>>>makes it deviate from being MSK?   
>
>>>Not at all.  It no longer meets the exact definition of MSK,
>>>once filtered.
>
>>We're in agreement there.
>
>>>>You seem to be claiming that filtering PSK makes it not PSK.   
>
>>>That is not exactly what I said.  I did say if one is choosing to
>>>say that filtering MSK can make it "not MSK", one can equally say that

>>>filtering PSK can make it "not PSK".
>
>>What you said was:
>
>>>>Eric: If you apply any kind of filtering to MSK, it's not MSK any
more.
>
>>>Steve: True.  The exact same could be said about PSK.   
>
>>The above is not a "can" or "may" statement.  This statement says that
>>any filtering changes PSK to be something other than PSK.   
>
>I said it "could be said" that filtering changes PSK into something other

>than PSK.  Saying "is could be said X" is not the same as saying "X".
>
>>This misses the fact that MSK and PSK are different in this regard.
>
>Well, I agree they are different in this regard, but not as different
>as you think.  
>
>MSK, by definition, has a pulse shape that is non-zero for exactly one 
>symbol.  PSK may or may not; but in those cases where it does
>(which is very common), adding additional filtering such that the
>overall filter response is longer than a symbol introduces A.M into
>the signal.  From this perspective MSK and PSK are behaving similarly
>under added filtering.
>
>> Filtering is necessary in PSK to achieve matched filter performance.  

>
>>What you said was:
>
>>>I will counter by saying that, in the general case, filtering 
>>>a phase-modulated signal will add amplitude modulation, thereby moving
it
>
>>>outside of the definition of being purely phase modulated.
>
>>That's not "if" it adds AM, you said it "will" add AM, and will
>>therefore no longer be PSK.  
>
>No, you are not paraphrasing me correctly, due to my having used
>the qualifying language "in the general case".  
>
>Consider:
>
>"In the general case, real numbers are not rational." (true statement)
>"real numbers are not rational"  (false statement)
>
>>Most matched filter pulse shapes used in PSK do add some AM, but in
>>the receiver it is between the symbol sampling instances and therefore
>>adds no AM to the "modulation" or the sliced constellation.   
>
>What counts in this discussion is the waveform of the signal being 
>transmitted.  Processing within the receiver doesn't count.
>
>>So, an infinite variety of filters can be selected which do not add AM
>>to the PSK modulation constellation in the receiver.   
>
>Sure.  But I have not been talking about signals deep within some
>receiver algorithm.  
>
>Suppose I construct an algorithm that takes a PAM signal and outputs
>an FSK signal; shall I then be able to claim that, by transmitting the 
>PAM signal, I was really transmitting FSK?
>
>Steve

Hi All,

I solved issues with frequency offset. Receiver is still noncoherent, but
I use double correlation on phase slope. Thank you all on comments. Now I
just need to translate matlab code to verilog ;).
---------------------------------------
Posted through http://www.DSPRelated.com

tarikkazaz <50642@DSPRelated> wrote:

>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>>
>Hi All,
>
>I solved issues with frequency offset. Receiver is still noncoherent, but
>I use double correlation on phase slope. Thank you all on comments. Now I
>just need to translate matlab code to verilog ;).

Great news!

Steve

On Wed, 7 Oct 2015 03:28:42 +0000 (UTC), spope33@speedymail.org (Steve
Pope) wrote:

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>On Tue, 6 Oct 2015 22:54:47 +0000 (UTC), spope33@speedymail.org (Steve
>
>>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>>>>Are you disagreeing that adding a filter that changes an MSK waveform
>>>>makes it deviate from being MSK?   
>
>>>Not at all.  It no longer meets the exact definition of MSK,
>>>once filtered.
>
>>We're in agreement there.
>
>>>>You seem to be claiming that filtering PSK makes it not PSK.   
>
>>>That is not exactly what I said.  I did say if one is choosing to
>>>say that filtering MSK can make it "not MSK", one can equally say that 
>>>filtering PSK can make it "not PSK".
>
>>What you said was:
>
>>>>Eric: If you apply any kind of filtering to MSK, it's not MSK any more.
>
>>>Steve: True.  The exact same could be said about PSK.   
>
>>The above is not a "can" or "may" statement.  This statement says that
>>any filtering changes PSK to be something other than PSK.   
>
>I said it "could be said" that filtering changes PSK into something other 
>than PSK.  Saying "is could be said X" is not the same as saying "X".
>
>>This misses the fact that MSK and PSK are different in this regard.
>
>Well, I agree they are different in this regard, but not as different
>as you think.  

It's clear to me that you don't know what I think.

>MSK, by definition, has a pulse shape that is non-zero for exactly one 
>symbol.  PSK may or may not; but in those cases where it does
>(which is very common), adding additional filtering such that the
>overall filter response is longer than a symbol introduces A.M into
>the signal.  From this perspective MSK and PSK are behaving similarly
>under added filtering.

Maybe this is what you're missing:   PSK has a non-constant envelope
(unlike MSK),  regardless of what you do, regardless of what filtering
you use, regardless of how long the pulse shape is or is not.   This
is because it has 180-degree phase transitions during which the
magnitude of the output vector collapses to ZERO.   This never, ever
happens with MSK (or any CPM waveform), which always have a constant
envelope and a PAPR = 1.

It doesn't matter, at all, how long the pulse shape is, there is
always a transition through the origin with ordinary PSK.
Offset-QPSK, on the other hand, introduces the channel offset
specifically to prevent that transition through zero so that the
output vector has a more constant magnitude.    If you restrict the
OQPSK pulse shape to a half-sine with one symbol direction, then it
happens to become equivalent to MSK.

If you use that same, one-symbol-long, half-sine pulse shape with QPSK
(i.e., non-offset QPSK), you get even more PAPR because the output
vector collapses to zero on every symbol transition rather than just
the 180-degree transitions.  So it really doesn't have much to do with
how long the pulse is,  the difference is really that QPSK has
transitions through the origin, as well as other possible transitions
that also add some (but not as much) PAPR.

Consider, for NRZ input and no filter, the output vector of an MSK
modulator produces a constant-envelope output, or a circle on the IQ
plane.   A QPSK modulator with NRZ input and no filter, produces a
square on the IQ plane with an X in it for the 180-degree transitions.
MSK has no AM, ever.   QPSK has AM, always.   It cannot be avoided.
Adding filtering that decreases the required channel width does
increase the PAPR, but it was already there to start with.

Here's an example:

http://image.slidesharecdn.com/ece414chapter3w12-141014045637-conversion-gate01/95/ece414-chapter3-w12-75-638.jpg?cb=1413262737

On the left is a QPSK signal showing the traces, including the X in
the middle for the 180-degree transitions.    Even the transitions on
the sides of the box don't trace a circle (like MSK does), because the
signal is quite different.  On the right is pi/4-QPSK, which, like
OQPSK, specifically reduces the PAPR associated with those 180-degree
transitions.   It still has a fair amount of PAPR, but it's less, and
if you want to use the type of PA that has to be constantly driven,
then you need to avoid those zero crossings.   pi/4-QPSK does that, so
does OQPSK, and they both do so with the same power efficiency as
QPSK.

The same diagram for MSK (or any CPM waveform, including TFM, GMSK,
CQPSK, etc., etc.) would be a circle, and just a circle, with no
deviation from that circle.   If it's not a circle, it's not CPM or
MSK.

PSK and MSK are very different in this regard.   Filtering does
increase the PAPR with PSK, but it is not the reason AM is there, and
it doesn't have that much to do with how long the pulse shape is.   It
has AM even with a pulse shape one symbol long.

To the original point, applying filtering to PSK that changes the
pulse shape doesn't make it something other than PSK, even if it adds
a lot of AM to the transmitted signal.    This is not true for MSK.

>> Filtering is necessary in PSK to achieve matched filter performance.   
>
>>What you said was:
>
>>>I will counter by saying that, in the general case, filtering 
>>>a phase-modulated signal will add amplitude modulation, thereby moving it 
>>>outside of the definition of being purely phase modulated.
>
>>That's not "if" it adds AM, you said it "will" add AM, and will
>>therefore no longer be PSK.  
>
>No, you are not paraphrasing me correctly, due to my having used
>the qualifying language "in the general case".  
>
>Consider:
>
>"In the general case, real numbers are not rational." (true statement)
>"real numbers are not rational"  (false statement)
>
>>Most matched filter pulse shapes used in PSK do add some AM, but in
>>the receiver it is between the symbol sampling instances and therefore
>>adds no AM to the "modulation" or the sliced constellation.   
>
>What counts in this discussion is the waveform of the signal being 
>transmitted.  Processing within the receiver doesn't count.

In a PSK system half of the pulse matched-filtering is typically done
in the receiver, so what's done in the receiver shouldn't be ignored.
It's part of the system design.

Regardless, the points I made above still hold true, even ignoring the
processing in the receiver.

>>So, an infinite variety of filters can be selected which do not add AM
>>to the PSK modulation constellation in the receiver.   
>
>Sure.  But I have not been talking about signals deep within some
>receiver algorithm.  

With PSK it is part of the system, since there is almost always
matched filtering before the information is detected.

>Suppose I construct an algorithm that takes a PAM signal and outputs
>an FSK signal; shall I then be able to claim that, by transmitting the 
>PAM signal, I was really transmitting FSK?

Doesn't seem germaine to this discussion.
Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

On Wed, 07 Oct 2015 06:41:46 -0500, "tarikkazaz" <50642@DSPRelated>
wrote:

>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>>
>>>On Tue, 6 Oct 2015 22:54:47 +0000 (UTC), spope33@speedymail.org (Steve
>>
>>>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>>
>>>>>Are you disagreeing that adding a filter that changes an MSK waveform
>>>>>makes it deviate from being MSK?   
>>
>>>>Not at all.  It no longer meets the exact definition of MSK,
>>>>once filtered.
>>
>>>We're in agreement there.
>>
>>>>>You seem to be claiming that filtering PSK makes it not PSK.   
>>
>>>>That is not exactly what I said.  I did say if one is choosing to
>>>>say that filtering MSK can make it "not MSK", one can equally say that
>
>>>>filtering PSK can make it "not PSK".
>>
>>>What you said was:
>>
>>>>>Eric: If you apply any kind of filtering to MSK, it's not MSK any
>more.
>>
>>>>Steve: True.  The exact same could be said about PSK.   
>>
>>>The above is not a "can" or "may" statement.  This statement says that
>>>any filtering changes PSK to be something other than PSK.   
>>
>>I said it "could be said" that filtering changes PSK into something other
>
>>than PSK.  Saying "is could be said X" is not the same as saying "X".
>>
>>>This misses the fact that MSK and PSK are different in this regard.
>>
>>Well, I agree they are different in this regard, but not as different
>>as you think.  
>>
>>MSK, by definition, has a pulse shape that is non-zero for exactly one 
>>symbol.  PSK may or may not; but in those cases where it does
>>(which is very common), adding additional filtering such that the
>>overall filter response is longer than a symbol introduces A.M into
>>the signal.  From this perspective MSK and PSK are behaving similarly
>>under added filtering.
>>
>>> Filtering is necessary in PSK to achieve matched filter performance.  
>
>>
>>>What you said was:
>>
>>>>I will counter by saying that, in the general case, filtering 
>>>>a phase-modulated signal will add amplitude modulation, thereby moving
>it
>>
>>>>outside of the definition of being purely phase modulated.
>>
>>>That's not "if" it adds AM, you said it "will" add AM, and will
>>>therefore no longer be PSK.  
>>
>>No, you are not paraphrasing me correctly, due to my having used
>>the qualifying language "in the general case".  
>>
>>Consider:
>>
>>"In the general case, real numbers are not rational." (true statement)
>>"real numbers are not rational"  (false statement)
>>
>>>Most matched filter pulse shapes used in PSK do add some AM, but in
>>>the receiver it is between the symbol sampling instances and therefore
>>>adds no AM to the "modulation" or the sliced constellation.   
>>
>>What counts in this discussion is the waveform of the signal being 
>>transmitted.  Processing within the receiver doesn't count.
>>
>>>So, an infinite variety of filters can be selected which do not add AM
>>>to the PSK modulation constellation in the receiver.   
>>
>>Sure.  But I have not been talking about signals deep within some
>>receiver algorithm.  
>>
>>Suppose I construct an algorithm that takes a PAM signal and outputs
>>an FSK signal; shall I then be able to claim that, by transmitting the 
>>PAM signal, I was really transmitting FSK?
>>
>>Steve
>
>Hi All,
>
>I solved issues with frequency offset. Receiver is still noncoherent, but
>I use double correlation on phase slope. Thank you all on comments. Now I
>just need to translate matlab code to verilog ;).

See what you started?   ;)

Glad to hear fo the progress.  Good luck with it.   


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

>On Wed, 7 Oct 2015 03:28:42 +0000 (UTC), spope33@speedymail.org (Steve

>>MSK, by definition, has a pulse shape that is non-zero for exactly one 
>>symbol.  PSK may or may not; but in those cases where it does
>>(which is very common), adding additional filtering such that the
>>overall filter response is longer than a symbol introduces A.M into
>>the signal.  From this perspective MSK and PSK are behaving similarly
>>under added filtering.

>Maybe this is what you're missing:   PSK has a non-constant envelope
>(unlike MSK),  regardless of what you do, regardless of what filtering
>you use, regardless of how long the pulse shape is or is not.   This
>is because it has 180-degree phase transitions during which the
>magnitude of the output vector collapses to ZERO.   This never, ever
>happens with MSK (or any CPM waveform), which always have a constant
>envelope and a PAPR = 1.

Yes, you've said this a couple times, and I have been thinking about it.

In my thinking, which I think is mainstream, MSK and PSK are both constant 
envelope, whereas MSK is continuous phase and PSK is not.

Do the phase discontinuities in PSK make it non-constant envelope?
Maybe.  Unfiltered, PSK has constant energy per baud, which is similar
to being constant envelope.  Apply any filtering, and the envelope
is now non-constant.  (Although, if the carrier frequency far exceeds
the baud rate, the effect is very small.)

So I see what you are saying, but I mostly see it as more evidence that
filtering (which is inevitable in reality) in fact moves PSK away from 
meeting the exact definition of PSK.

[snip]

>PSK and MSK are very different in this regard.   

I agree they are different.

> Filtering does increase the PAPR with PSK, but it is not the
> reason AM is there, and it doesn't have that much to do with
> how long the pulse shape is.   It has AM even with a pulse
> shape one symbol long.

You've said this last sentence a couple times too, and I can't
really agree with it.  PSK with a rectangular pulse shape one 
symbol long has constant energy per baud

>
>To the original point, applying filtering to PSK that changes the
>pulse shape doesn't make it something other than PSK, even if it adds
>a lot of AM to the transmitted signal.    This is not true for MSK.
>
>>> Filtering is necessary in PSK to achieve matched filter performance.   
>>
>>>What you said was:
>>
>>>>I will counter by saying that, in the general case, filtering 
>>>>a phase-modulated signal will add amplitude modulation, thereby moving it 
>>>>outside of the definition of being purely phase modulated.
>>
>>>That's not "if" it adds AM, you said it "will" add AM, and will
>>>therefore no longer be PSK.  
>>
>>No, you are not paraphrasing me correctly, due to my having used
>>the qualifying language "in the general case".  
>>
>>Consider:
>>
>>"In the general case, real numbers are not rational." (true statement)
>>"real numbers are not rational"  (false statement)
>>
>>>Most matched filter pulse shapes used in PSK do add some AM, but in
>>>the receiver it is between the symbol sampling instances and therefore
>>>adds no AM to the "modulation" or the sliced constellation.   
>>
>>What counts in this discussion is the waveform of the signal being 
>>transmitted.  Processing within the receiver doesn't count.
>
>In a PSK system half of the pulse matched-filtering is typically done
>in the receiver, so what's done in the receiver shouldn't be ignored.
>It's part of the system design.
>
>Regardless, the points I made above still hold true, even ignoring the
>processing in the receiver.
>
>>>So, an infinite variety of filters can be selected which do not add AM
>>>to the PSK modulation constellation in the receiver.   
>>
>>Sure.  But I have not been talking about signals deep within some
>>receiver algorithm.  
>
>With PSK it is part of the system, since there is almost always
>matched filtering before the information is detected.
>
>>Suppose I construct an algorithm that takes a PAM signal and outputs
>>an FSK signal; shall I then be able to claim that, by transmitting the 
>>PAM signal, I was really transmitting FSK?
>
>Doesn't seem germaine to this discussion.
>Eric Jacobsen
>Anchor Hill Communications
>http://www.anchorhill.com

Steve Pope <spope33@speedymail.org> wrote:

>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:

[ I hit send before I was done editing.  To continue: ]

>>PSK and MSK are very different in this regard.   
>
>I agree they are different.
>
>> Filtering does increase the PAPR with PSK, but it is not the
>> reason AM is there, and it doesn't have that much to do with
>> how long the pulse shape is.   It has AM even with a pulse
>> shape one symbol long.

>You've said this last sentence a couple times too, and I can't
>really agree with it.  PSK with a rectangular pulse shape one 
>symbol long has constant energy per baud.

A.M. is usually considered to be amplitude modulation of the envelope of 
the carrier, and I do not see how PSK with the above pulse shape 
has any A.M.

Nor do I see where the PAPR of this signal is different from
that of an unmodulated carrier.  Unless there's further filtering.

>>>What counts in this discussion is the waveform of the signal being 
>>>transmitted.  Processing within the receiver doesn't count.

>>In a PSK system half of the pulse matched-filtering is typically done
>>in the receiver, so what's done in the receiver shouldn't be ignored.
>>It's part of the system design.

I'm not saying to ignore receiver processing on a global basis;
but when answering questions of the form, "What type(s) of modulation
does the transmitted wavefore exhibit" (which is I believe the
topic at hand), what's done in the receiver is not relevant.

Regardless of the above, to me, "PSK with a little AM" is still PSK,
in normal discussions of the matter, in most contexts, unless
one is hewing very strictly to exact definitions.


Steve

On Wed, 7 Oct 2015 23:34:02 +0000 (UTC), spope33@speedymail.org (Steve
Pope) wrote:

>Steve Pope <spope33@speedymail.org> wrote:
>
>>Eric Jacobsen <eric.jacobsen@ieee.org> wrote:
>
>[ I hit send before I was done editing.  To continue: ]
>
>>>PSK and MSK are very different in this regard.   
>>
>>I agree they are different.
>>
>>> Filtering does increase the PAPR with PSK, but it is not the
>>> reason AM is there, and it doesn't have that much to do with
>>> how long the pulse shape is.   It has AM even with a pulse
>>> shape one symbol long.
>
>>You've said this last sentence a couple times too, and I can't
>>really agree with it.  PSK with a rectangular pulse shape one 
>>symbol long has constant energy per baud.
>
>A.M. is usually considered to be amplitude modulation of the envelope of 
>the carrier, and I do not see how PSK with the above pulse shape 
>has any A.M.

>Nor do I see where the PAPR of this signal is different from
>that of an unmodulated carrier.  Unless there's further filtering.

Do you think that PSK is only PSK in the strict condition that it only
has rectangular pulses?    Is it still PSK if there is a matched
filter, say an RRC, on both ends of the link?  

I'll give you a hint:   It is still PSK with matched filters, and once
there's a filter on the rectangular pulse there is a big amplitude
change around the 180 degree transitions.    It's there in any
practical system even with rectangular pulses, too, but the region of
the amplitude change is much smaller in time and pedantic nits will be
picked that it isn't really there in the theoretical NRZ case since
the time spread goes to zero, as you just mentioned.

Regardless, you cannot escape that PSK is still PSK even with matched
filters, in which case there is lots of amplitude changes in the
signal, but not in the received constellation after the matching
filter.   Still PSK, always has been.

Filtering PSK does not make it not PSK.   PSK is nearly always
filtered, even in the textbooks that describe how to do it.  PSK will
have AM, especially between the symbols, and it is, despite your claim
to the contrary, still PSK.

>>>>What counts in this discussion is the waveform of the signal being 
>>>>transmitted.  Processing within the receiver doesn't count.
>
>>>In a PSK system half of the pulse matched-filtering is typically done
>>>in the receiver, so what's done in the receiver shouldn't be ignored.
>>>It's part of the system design.
>
>I'm not saying to ignore receiver processing on a global basis;
>but when answering questions of the form, "What type(s) of modulation
>does the transmitted wavefore exhibit" (which is I believe the
>topic at hand), what's done in the receiver is not relevant.

Whether there's AM on the constellation, i.e., the modulation where
the information is contained, is only discernible inside the
demodulator after the matched filter.    So what's done inside the
demodulator is, in fact, relevant.

But, again, it isn't necessary to look inside the demodulator to see
that PSK has a lot of AM during 180-degree transitions.   It's hard
not to see it, actually.   Any inspection of an eye diagram shows it
pretty clearly.   I already gave a link to example eye diagrams with
different filtering to show how the AM changes as the filter narrows.
The crossing points of the eyes where the amplitude goes to zero is
obvious.

In case you missed it (Figure 7 and Table 1):

http://www.dsprelated.com/showarticle/60.php  

>Regardless of the above, to me, "PSK with a little AM" is still PSK,
>in normal discussions of the matter, in most contexts, unless
>one is hewing very strictly to exact definitions.

PSK will have more and more PAPR (aka AM)  the narrower it is
filtered.   Unlike MSK or any other CPM, this does not cost power
efficiency with PSK, nor does it cause it to cease to be PSK.   MSK,
or any CPM, does not have the 180-degree transitions that PSK has.

Filtering MSK makes it not MSK.   PSK is still PSK even after the
application of pulse-shaping filters, even if those pulse shaping
filters increase the PAPR (AM) of PSK.

  
Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com