| Author |
Message |
john
Guest
|
 Posted:
Thu Jan 06, 2005 4:38 pm Post subject:
Processing Gain after Integrating DPSK Demod Output |
|
|
Hello group,
I have a question about how to compute processing gain from video
integration at the output of a noncoherent PSK demodulator.
Suppose I have a differentially encoded, BPSK-modulated data message of
length N bits that repeats M times in a row. To detect the message, I
apply the modulated data to a delay-and-multiply followed by a one-bit
boxcar filter. Next I "stack and add" all M repetitions of the data in
a circular buffer that contains N bits worth of samples -- video
integration. After all M reps have been accumulated I go find the bit
centers and check for a good message (all bits must be correct).
The normal error curve for DPSK is BER = exp(-Eb/No)/2. If M=1, I would
expect the message error rate to be MER = 1-(1-BER)^N.
If M > 1, can I compute a processing gain term P, and boost Eb/No by
that amount to predict MER? If so, what is P? I assume it lies
somewhere between 0 and 10*log10(M).
Thanks!
John |
|
| Back to top |
|
 |
Eric Jacobsen
Guest
|
| Posted:
Fri Jan 07, 2005 3:26 am Post subject:
Re: Processing Gain after Integrating DPSK Demod Output |
|
|
On 6 Jan 2005 03:38:20 -0800, "john" <johns@xetron.com> wrote:
| Quote: | Hello group,
I have a question about how to compute processing gain from video
integration at the output of a noncoherent PSK demodulator.
Suppose I have a differentially encoded, BPSK-modulated data message of
length N bits that repeats M times in a row. To detect the message, I
apply the modulated data to a delay-and-multiply followed by a one-bit
boxcar filter. Next I "stack and add" all M repetitions of the data in
a circular buffer that contains N bits worth of samples -- video
integration. After all M reps have been accumulated I go find the bit
centers and check for a good message (all bits must be correct).
The normal error curve for DPSK is BER = exp(-Eb/No)/2. If M=1, I would
expect the message error rate to be MER = 1-(1-BER)^N.
If M > 1, can I compute a processing gain term P, and boost Eb/No by
that amount to predict MER? If so, what is P? I assume it lies
somewhere between 0 and 10*log10(M).
Thanks!
John
|
If you're trying to find the effective gain in Packet Error Rate,
Frame Error Rate, or whatever your preferred metric may be, due to the
M-fold integration, you might just compare both ways and plot the
difference.
Do I understand correctly that you're just collecting M packets and
adding them together before decoding (i.e., Chase combining)?
A hopefully easy thing to do would be to turn off the M-fold
integration and plot the packet (or frame, whatever you want to call
them) versus SNR. Then turn it on and do it again. The difference
in the SNRs at some representative and meaningful Packet Error Rate is
the processing gain wrt SNR.
Repetition coding is typically not very efficient, so if you did the
same thing vs Eb/No instead of SNR you'd likely not see much gain. If
the system is nicely functional consuming bandwidth with M-fold
repetitions then that's fine, but if you ever get constrained for
bandwidth you might want to look at other methods.
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org |
|
| Back to top |
|
|
john
Guest
|
| Posted:
Fri Jan 07, 2005 3:45 am Post subject:
Re: Processing Gain after Integrating DPSK Demod Output |
|
|
Eric Jacobsen wrote:
| Quote: | On 6 Jan 2005 03:38:20 -0800, "john" <johns@xetron.com> wrote:
Hello group,
I have a question about how to compute processing gain from video
integration at the output of a noncoherent PSK demodulator.
Suppose I have a differentially encoded, BPSK-modulated data message
of
length N bits that repeats M times in a row. To detect the message,
I
apply the modulated data to a delay-and-multiply followed by a
one-bit
boxcar filter. Next I "stack and add" all M repetitions of the data
in
a circular buffer that contains N bits worth of samples -- video
integration. After all M reps have been accumulated I go find the
bit
centers and check for a good message (all bits must be correct).
The normal error curve for DPSK is BER = exp(-Eb/No)/2. If M=1, I
would
expect the message error rate to be MER = 1-(1-BER)^N.
If M > 1, can I compute a processing gain term P, and boost Eb/No by
that amount to predict MER? If so, what is P? I assume it lies
somewhere between 0 and 10*log10(M).
Thanks!
John
If you're trying to find the effective gain in Packet Error Rate,
Frame Error Rate, or whatever your preferred metric may be, due to
the
M-fold integration, you might just compare both ways and plot the
difference.
Do I understand correctly that you're just collecting M packets and
adding them together before decoding (i.e., Chase combining)?
A hopefully easy thing to do would be to turn off the M-fold
integration and plot the packet (or frame, whatever you want to call
them) versus SNR. Then turn it on and do it again. The difference
in the SNRs at some representative and meaningful Packet Error Rate
is
the processing gain wrt SNR.
Repetition coding is typically not very efficient, so if you did the
same thing vs Eb/No instead of SNR you'd likely not see much gain.
If
the system is nicely functional consuming bandwidth with M-fold
repetitions then that's fine, but if you ever get constrained for
bandwidth you might want to look at other methods.
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org
|
Thanks, Eric.
I am collecting M packets worth of samples together (multiple samples
per bit), adding them up, then finding bit centers and decoding. So the
addition is done on 16-bit samples, not hard decisions.
I agree that this is not an ideal scheme, but it is a legacy waveform
and I only control the receiver.
I am after some kind of expected processing gain, to compare to
measured data. The analysis would have to take into account the
non-gaussian PDF of the noise, since it goes through a delay and
multiply circuit before integration. This morning I was thinking that I
could accumulate a bunch of gaussian * gaussian data and see how much
the variance reduces as a function of the number accumulated.
Thanks again,
John |
|
| Back to top |
|
|
Eric Jacobsen
Guest
|
| Posted:
Fri Jan 07, 2005 11:49 pm Post subject:
Re: Processing Gain after Integrating DPSK Demod Output |
|
|
On 6 Jan 2005 14:45:11 -0800, "john" <johns@xetron.com> wrote:
| Quote: | I am collecting M packets worth of samples together (multiple samples
per bit), adding them up, then finding bit centers and decoding. So the
addition is done on 16-bit samples, not hard decisions.
I agree that this is not an ideal scheme, but it is a legacy waveform
and I only control the receiver.
|
So you're adding them up prior to synchronization? That will muddy
up an analysis since now the ISI due to the possible synchronization
errors before addition will also be a random variable with some
non-friendly analytical behavior.
Is there a way to synchronize them first and then add them coherently?
This is essentially Chase combining and is pretty well known and
understood. If the channels are gaussian then the gain is pretty
straightforward but if there's fading I think a literature stuff would
turn up a lot of treatment on that. If you channel isn't a common
one it might be harder to find something.
| Quote: | I am after some kind of expected processing gain, to compare to
measured data. The analysis would have to take into account the
non-gaussian PDF of the noise, since it goes through a delay and
multiply circuit before integration. This morning I was thinking that I
could accumulate a bunch of gaussian * gaussian data and see how much
the variance reduces as a function of the number accumulated.
Thanks again,
|
Oh, so you're just trying to see whether it is working as well as
should be expected? You're right that this is potentially quite
non-trivial. Duzzat help? :(
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in