[need help] amplifier and filter for sharp impluse signal

Do you mean that the channel behaves as a differentiator (CR circuit) ? Maybe some other baseband mode other than NRZ would improve SNR. ( i.e : a symetric signal will give a better SNR ) Can you tell us more about this communication channel ?

Did you mean *Hysteresis* Comparator ?

Ok, any filter reducing the noise will reduce the energy of the impulse as well. However, if the frequency of the clock of the NRZ signal is well defined and stable, you can try to bandpass the received signal to slightly improve the SNR. Is the noise distribution equiprobable, or does it have some shape ?

BTW, depending on your transfer speed and channel bandpass, you may consider to use some error-correction code to reduce transmission errors due to a low SNR.

Does a zero/one NRZ transition be received as a positive impulse, and a one/zero transition be received as a negative impulse ? In this case, you can use a comparator followed by a flip-flop to rebuild the original NRZ signal : the positive pulse will SET the Q output, the negative pulse will RESET the Q output, ans the signal at Q is the original NRZ. ( I hope I made it clear enough )

"... because the spectrum of the noise in our system is close to the impulse".

This is going to be a problem for you, for sure. The most basic role of the detector in a communications system is to tell the difference between noise and signal. You can kill yourself trying to flog a system whose signal looks like noise, or you can change the signal to make your communications channel easier to deal with.

If you have the freedom to change the signal, you should do so. Even if you can't change the NRZ character of the signal, consider whether you can add some forward error correction to it, so that the system as a whole is robust against high bit errors in the raw reception.

If you don't, then start thinking about _everything_ that distinguishes signal from noise. In your case, it sounds like you have fairly white noise, and a signal that's the derivative of a square pulse. So if your pulse widths are constant, you should be able to get at least a 3dB gain in SNR with a matched filter that correlates a downward impulse with a previous upward impulse. Assuming you get the channel characteristic right, you may be surprised at just how much more signal you pull out of the noise.

--
http://www.wescottdesign.com

Schmitt trigger.

Schmidt is beer.

Hope This Helps! Rich

yes! like a differential circuit

yes, I mean Hysteresis Comparator ? Sorry for my English expression.

the noise spectrum is time-varying. some times the noise is low, the other times, it distributes equiprobably.

Thanks for your advice. However, I consider to optimize the PHY design firstly because the noise problem is really a trouble.

It sounds like your channel doesn't have enough low frequency response to pass the square wave, so you get impulses. You would be better off shaping the signal to fit the channel bandwidth.

Basically, your project sounds hairbrained to me. If the signal looks like the noise, demod will not be poor.

thanks for your reply. the channel is capacitive and this characteristic can not be changed.

do you mean that we should use modulation and demodulation?

Sound strange to me, and interresting too ... It would help a lot if you could specify details on the physics of the comm channel.

This is indeed the trickiest part : if the noise distribution is not stable, ( i.e : variable noise level vs frequency band ) then the SNR will be variable - and so the error rate.

There is a way to improve such situation, called "spread spectrum" which enlarges the signal spectrum over the channel bandwidth, this gives a large redundancy to overcome severe interference, and greatly helps to lower the error probability.

Another technique is used for channels having time-variant impulse responses, which is a consequence of the constantly changing physical characteristics of the channel.

Does it sounds like the problem you are facing ?

Yes, whatever will be the choice, you still keep the option to improve the overall transfer by implementing a convenient error detection/correction code at the software level. However I doubt that NRZ is a good choice in this tricky case, you may have to use a convenient modulation w/ matched filters.

Anyway such subjects are well covered - in depth - in books that I strongly suggest you to read :

1- "Digital Communications - John Proakis" (the Bible) 2- "Digital And Analog Communication Systems - Leon Couch"

HTH

thanks for your reply. Things have been made a little progress. The SNR is improved however, the amplitude of signal and noise change together with some other factors. I will put the pictures of print screen of the scope on a website as soon as possible.