Forums

CPM

Started by Vladimir Vassilevsky April 25, 2010
I was trying to design a bandlimited pseudorandom signal with constant 
envelope for radar/sonar and like applications. I.e. frequency 
modulation with smoothed transitions. The problem is closely related to 
CPM for data transmission, however the optimization criteria are not 
quite the same. In particular, it turns out that optimal value for 
normalized deviation is higher then 0.5, optimal h ~0.6..0.7.
Another observation: the binary CPM signal with BT ~ 0.3 and  h ~ 0.8 
makes for the best possible (?) rectangular spectrum shape, however its 
autocorrelation properties are worse then optimal.

Could you suggest a good reading on this matter?

Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com
On Apr 25, 8:28&#2013266080;pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> I was trying to design a bandlimited pseudorandom signal with constant > envelope for radar/sonar and like applications. I.e. frequency > modulation with smoothed transitions. The problem is closely related to > CPM for data transmission, however the optimization criteria are not > quite the same. In particular, it turns out that optimal value for > normalized deviation is higher then 0.5, optimal h ~0.6..0.7. > Another observation: the binary CPM signal with BT ~ 0.3 and &#2013266080;h ~ 0.8 > makes for the best possible (?) rectangular spectrum shape, however its > autocorrelation properties are worse then optimal. > > Could you suggest a good reading on this matter? > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com
http://books.google.com/books?id=RwMpVSOXdJQC&printsec=frontcover&dq=digital+phase+modulation&source=bl&ots=7qKVpiwre8&sig=kqxvTHpdWkgcWYxNryZSxyaSk8c&hl=en&ei=k-PUS6uMEYT48AbgssSqDw&sa=X&oi=book_result&ct=result&resnum=4&ved=0CB4Q6AEwAw#v=onepage&q&f=false John
On Apr 25, 8:28&#2013266080;pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> I was trying to design a bandlimited pseudorandom signal with constant > envelope for radar/sonar and like applications. I.e. frequency > modulation with smoothed transitions. The problem is closely related to > CPM for data transmission, however the optimization criteria are not > quite the same. In particular, it turns out that optimal value for > normalized deviation is higher then 0.5, optimal h ~0.6..0.7. > Another observation: the binary CPM signal with BT ~ 0.3 and &#2013266080;h ~ 0.8 > makes for the best possible (?) rectangular spectrum shape, however its > autocorrelation properties are worse then optimal. > > Could you suggest a good reading on this matter? > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com
BTW, I came to the conclusion on a prior project that CPM cannot have impulse-like autocorrelation. For my project I switched to SQPSK with a PN sequence (different on I and Q) and accepted suboptimal peak/avg ratio in exchange for ideal autocorrelation. John John
Vladimir Vassilevsky  <nospam@nowhere.com> wrote:

>I was trying to design a bandlimited pseudorandom signal with constant >envelope for radar/sonar and like applications. I.e. frequency >modulation with smoothed transitions. The problem is closely related to >CPM for data transmission, however the optimization criteria are not >quite the same. In particular, it turns out that optimal value for >normalized deviation is higher then 0.5, optimal h ~0.6..0.7. >Another observation: the binary CPM signal with BT ~ 0.3 and h ~ 0.8 >makes for the best possible (?) rectangular spectrum shape, however its >autocorrelation properties are worse then optimal. > >Could you suggest a good reading on this matter?
I have no idea as to literature, but I likewise have had trouble sometimes receiving an h = 0.5 signal as well as I would like or would have expected, and had to push to a higher value of h. I assume it's simply because I don't know the best algorithms for what I'm trying to do. Steve

John wrote:
> On Apr 25, 8:28 pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote: > >>I was trying to design a bandlimited pseudorandom signal with constant >>envelope for radar/sonar and like applications. I.e. frequency >>modulation with smoothed transitions. The problem is closely related to >>CPM for data transmission, however the optimization criteria are not >>quite the same. In particular, it turns out that optimal value for >>normalized deviation is higher then 0.5, optimal h ~0.6..0.7. >>Another observation: the binary CPM signal with BT ~ 0.3 and h ~ 0.8 >>makes for the best possible (?) rectangular spectrum shape, however its >>autocorrelation properties are worse then optimal.
> BTW, I came to the conclusion on a prior project that CPM cannot have > impulse-like autocorrelation.
Can you clarify? I am getting the main autocorrelation lobe just like it supposed to be according to the bandwidth, however there is a second lobe at ~ -7dB, a third at -20..-30 dB and so on till it gets to the autocorrelation floor.
> For my project I switched to SQPSK with > a PN sequence (different on I and Q) and accepted suboptimal peak/avg > ratio in exchange for ideal autocorrelation.
I thought about it. With or without offset, filtered QPSK has near gaussian amplitude distribution. It is not very obvious if better autocorrelation will compensate for PAPR. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
> > John > > John

Steve Pope wrote:
> Vladimir Vassilevsky <nospam@nowhere.com> wrote: > > >>I was trying to design a bandlimited pseudorandom signal with constant >>envelope for radar/sonar and like applications. I.e. frequency >>modulation with smoothed transitions. The problem is closely related to >>CPM for data transmission, however the optimization criteria are not >>quite the same. In particular, it turns out that optimal value for >>normalized deviation is higher then 0.5, optimal h ~0.6..0.7. >>Another observation: the binary CPM signal with BT ~ 0.3 and h ~ 0.8 >>makes for the best possible (?) rectangular spectrum shape, however its >>autocorrelation properties are worse then optimal. >> >>Could you suggest a good reading on this matter? > > > I have no idea as to literature, but I likewise have had > trouble sometimes receiving an h = 0.5 signal as well as > I would like or would have expected, and had to push to > a higher value of h. > > I assume it's simply because I don't know the best algorithms > for what I'm trying to do.
For partial response CPM, the value of h somewhat higher then 0.5 makes for lower autocorrelation floor. The difference could be ~1..2dB. The intuitive explanation is if you choose h higher then 0.5, then the "average" signal is going to be closer to h=0.5 . Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
On Apr 25, 8:28&#2013266080;pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> I was trying to design a bandlimited pseudorandom signal with constant > envelope for radar/sonar and like applications. I.e. frequency > modulation with smoothed transitions. The problem is closely related to > CPM for data transmission, however the optimization criteria are not > quite the same. In particular, it turns out that optimal value for > normalized deviation is higher then 0.5, optimal h ~0.6..0.7. > Another observation: the binary CPM signal with BT ~ 0.3 and &#2013266080;h ~ 0.8 > makes for the best possible (?) rectangular spectrum shape, however its > autocorrelation properties are worse then optimal. > > Could you suggest a good reading on this matter? > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com
Try http://people.eecs.ku.edu/~esp/publications/c2009RadarCpm.pdf and references therein. col
> >Could you suggest a good reading on this matter? >
I don't know if the following will address your questions directly. http://people.eecs.ku.edu/~esp/doku/doku.php?id=publications But Dr. Perrins has his IEEE papers available from his webpage.
On Apr 25, 10:07&#2013266080;pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote:
> John wrote: > > On Apr 25, 8:28 pm, Vladimir Vassilevsky <nos...@nowhere.com> wrote: > > >>I was trying to design a bandlimited pseudorandom signal with constant > >>envelope for radar/sonar and like applications. I.e. frequency > >>modulation with smoothed transitions. The problem is closely related to > >>CPM for data transmission, however the optimization criteria are not > >>quite the same. In particular, it turns out that optimal value for > >>normalized deviation is higher then 0.5, optimal h ~0.6..0.7. > >>Another observation: the binary CPM signal with BT ~ 0.3 and &#2013266080;h ~ 0.8 > >>makes for the best possible (?) rectangular spectrum shape, however its > >>autocorrelation properties are worse then optimal. > > BTW, I came to the conclusion on a prior project that CPM cannot have > > impulse-like autocorrelation. > > Can you clarify? > I am getting the main autocorrelation lobe just like it supposed to be > according to the bandwidth, however there is a second lobe at ~ -7dB, a > third at -20..-30 dB and so on till it gets to the autocorrelation floor. > > > For my project I switched to SQPSK with > > a PN sequence (different on I and Q) and accepted suboptimal peak/avg > > ratio in exchange for ideal autocorrelation. > > I thought about it. With or without offset, filtered QPSK has near > gaussian amplitude distribution. It is not very obvious if better > autocorrelation will compensate for PAPR. > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultanthttp://www.abvolt.com > > > > > John > > > John
If you can live with the spectrum of class C amp driven with a non- constant env signal such as square-pulse offset QPSK, then I would think the autocorrelation and tx efficiency will be quite good. If the spectrum is unacceptable then you can either back off the transmit power (very costly), reduce the symbol rate (longer transmit time), or change the waveform to something with autocorrelation sidelobes such as CPM. It would take some experimentation or a good simulation to figure out which is better from an end-to-end link budget standpoint. John John