Forums

PPM question

Started by John December 28, 2007
Hi group,

I'm looking for new ideas on the topic of efficiently determining a
suitable slicing threshold for a PPM signal. What I have is a signal
containing short pulses in white noise. The pulse widths and
repetition intervals can vary depending on the specific PPM signal.
What I have tried is a long term average, as well as a block sorting
procedure that extracts a percentile value near 90%, so it excludes
the pulses. The long term average is biased if the pulses are too long
or too frequent. The sorting procedure is compute intensive and needs
to know some signal specifics.

Are there any other ideas for finding the noise floor in this
situation? It needs to be very efficient, working on blocks of about
32k samples at a time.

Thanks, and Happy New Year.

John
"John" <sampson164@gmail.com> wrote in message
news:c97b10bc-1311-4752-aa3d-a7546f5a9604@e10g2000prf.googlegroups.com...
> Hi group, > > I'm looking for new ideas on the topic of efficiently determining a > suitable slicing threshold for a PPM signal. What I have is a signal > containing short pulses in white noise. The pulse widths and > repetition intervals can vary depending on the specific PPM signal. > What I have tried is a long term average, as well as a block sorting > procedure that extracts a percentile value near 90%, so it excludes > the pulses. The long term average is biased if the pulses are too long > or too frequent. The sorting procedure is compute intensive and needs > to know some signal specifics. > > Are there any other ideas for finding the noise floor in this > situation? It needs to be very efficient, working on blocks of about > 32k samples at a time.
The optimal receiver for PAM or PPM is not very simple. You have to use the decision directed algorithm to set the threshold. Furthermore, since the noise statistics is different for '0' and '1' states, the Euclidean distance is not the optimal metrics. First, remove the DC and establish the bit synchronization. Then, estimate the levels of '1' and '0' of the original signal. This gives you E/No also. Then, estimate the threshold. Vladimir Vassilevsky DSP and Mixed Signal Consultant www.abvolt.com
On Dec 29, 1:14 am, "Vladimir Vassilevsky"
<antispam_bo...@hotmail.com> wrote:
> "John" <sampson...@gmail.com> wrote in message > > news:c97b10bc-1311-4752-aa3d-a7546f5a9604@e10g2000prf.googlegroups.com... > > > Hi group, > > > I'm looking for new ideas on the topic of efficiently determining a > > suitable slicing threshold for a PPM signal. What I have is a signal > > containing short pulses in white noise. The pulse widths and > > repetition intervals can vary depending on the specific PPM signal. > > What I have tried is a long term average, as well as a block sorting > > procedure that extracts a percentile value near 90%, so it excludes > > the pulses. The long term average is biased if the pulses are too long > > or too frequent. The sorting procedure is compute intensive and needs > > to know some signal specifics. > > > Are there any other ideas for finding the noise floor in this > > situation? It needs to be very efficient, working on blocks of about > > 32k samples at a time. > > The optimal receiver for PAM or PPM is not very simple. You have to use the > decision directed algorithm to set the threshold. Furthermore, since the > noise statistics is different for '0' and '1' states, the Euclidean distance > is not the optimal metrics. > > First, remove the DC and establish the bit synchronization. Then, estimate > the levels of '1' and '0' of the original signal. This gives you E/No also. > Then, estimate the threshold. > > Vladimir Vassilevsky > DSP and Mixed Signal Consultantwww.abvolt.com
Thank you, and happy new year. John
On Dec 29 2007, 10:09 pm, John <sampson...@gmail.com> wrote:
> On Dec 29, 1:14 am, "Vladimir Vassilevsky" > > > > <antispam_bo...@hotmail.com> wrote: > > "John" <sampson...@gmail.com> wrote in message > > >news:c97b10bc-1311-4752-aa3d-a7546f5a9604@e10g2000prf.googlegroups.com... > > > > Hi group, > > > > I'm looking for new ideas on the topic of efficiently determining a > > > suitable slicing threshold for aPPMsignal. What I have is a signal > > > containing short pulses in white noise. The pulse widths and > > > repetition intervals can vary depending on the specificPPMsignal. > > > What I have tried is a long term average, as well as a block sorting > > > procedure that extracts a percentile value near 90%, so it excludes > > > the pulses. The long term average is biased if the pulses are too long > > > or too frequent. The sorting procedure is compute intensive and needs > > > to know some signal specifics. > > > > Are there any other ideas for finding the noise floor in this > > > situation? It needs to be very efficient, working on blocks of about > > > 32k samples at a time. > > > The optimal receiver for PAM orPPMis not very simple. You have to use the > > decision directed algorithm to set the threshold. Furthermore, since the > > noise statistics is different for '0' and '1' states, the Euclidean distance > > is not the optimal metrics. > > > First, remove the DC and establish the bit synchronization. Then, estimate > > the levels of '1' and '0' of the original signal. This gives you E/No also. > > Then, estimate the threshold. > > > Vladimir Vassilevsky > > DSP and Mixed Signal Consultantwww.abvolt.com > > Thank you, and happy new year. > > John
John, I'm working on a similar project. My processor makes a threshold calculation for pulsed signals of widely varying duty cycles and amplitudes. Like yours, it calculates a long-term average to determine the noise floor. It works OK in most conditions, but I'm currently researching different techniques to improve it. What technique are you now using? Darol Klawetter
On Jan 17, 3:13 pm, Darol Klawetter <darol.klawet...@l-3com.com>
wrote:
> On Dec 29 2007, 10:09 pm, John <sampson...@gmail.com> wrote: > > > > > On Dec 29, 1:14 am, "Vladimir Vassilevsky" > > > <antispam_bo...@hotmail.com> wrote: > > > "John" <sampson...@gmail.com> wrote in message > > > >news:c97b10bc-1311-4752-aa3d-a7546f5a9604@e10g2000prf.googlegroups.com... > > > > > Hi group, > > > > > I'm looking for new ideas on the topic of efficiently determining a > > > > suitable slicing threshold for aPPMsignal. What I have is a signal > > > > containing short pulses in white noise. The pulse widths and > > > > repetition intervals can vary depending on the specificPPMsignal. > > > > What I have tried is a long term average, as well as a block sorting > > > > procedure that extracts a percentile value near 90%, so it excludes > > > > the pulses. The long term average is biased if the pulses are too long > > > > or too frequent. The sorting procedure is compute intensive and needs > > > > to know some signal specifics. > > > > > Are there any other ideas for finding the noise floor in this > > > > situation? It needs to be very efficient, working on blocks of about > > > > 32k samples at a time. > > > > The optimal receiver for PAM orPPMis not very simple. You have to use the > > > decision directed algorithm to set the threshold. Furthermore, since the > > > noise statistics is different for '0' and '1' states, the Euclidean distance > > > is not the optimal metrics. > > > > First, remove the DC and establish the bit synchronization. Then, estimate > > > the levels of '1' and '0' of the original signal. This gives you E/No also. > > > Then, estimate the threshold. > > > > Vladimir Vassilevsky > > > DSP and Mixed Signal Consultantwww.abvolt.com > > > Thank you, and happy new year. > > > John > > John, > > I'm working on a similar project. My processor makes a threshold > calculation for pulsed signals of widely varying duty cycles and > amplitudes. Like yours, it calculates a long-term average to determine > the noise floor. It works OK in most conditions, but I'm currently > researching different techniques to improve it. What technique are you > now using? > > Darol Klawetter
We've tried a few different ideas with varying results -- better in some conditions, worse in others. Nothing that is solidly better than the long term average. John