Hi folks,

I have ta analyse a signal transmitted in a burst mode in UHF and I have no apriori about it.
The signal seams to be transmitted using a basic 2-FSK modulation.

I have demodulated this signal as follows :
1) Downconvert the signal to O Hz. Now I have a complex base band FSK signal, oversampled by 10.
2) Apply a low pass filter in order to remove the noise. The cutoff frequency is set to be just above the frequency tone of the 2-FSK tone. The number a taps is lower than the FSK time symbol in order to avoid any intersymbol interference.
3) Perform an FM demodulation observing the angle between two consecutive samples (still oversampled by 10)
4) (optional) Apply a low pass filter in order to smooth the signal (frequency representation of the 2 FSK). Once again, The number a taps is lower than the FSK time symbol in order to avoid any intersymbol interference.

I am pretty confident that the signal is 2-FSK since
- the Power Spectral Density looks like a 2-FSK
- Sharp transitions after steps 3) and 4) with 2 states corresponding to the 2 frequency FSK tones

The cyclostationnarity of the signal after 3) or 4) gives a symbol rate of exactly half the frequency spacing between the 2 FSK tones.

My problem is the following. When I plot the eye diagram of the signal after 3) or 4) with a symbol rate of half the frequency spacing, I observe:
- a very clean eye diagram during tens (between 32 and up to 400) of bits/symbols => optimal decision is possible
- a transition in the eye diagram, that lasts about 4 bits/symbols
- a very clean eye diagram during tens (between 32 and up to 400) of bits/symbols => optimal decision is possible, but is different from the previous one
- a transition in the eye diagram, that lasts about 4 bits/symbols
- and so on...

Note that the bust duration is about 1024 symbols.

I have never seen such behaviour and I have absolutly no idea of what kind of signal it is how the signal is exactly modulated.

Does anyone have an idea on that? Is there any standard or system that could lead to this type of behaviour?

Thanks a lot!!!

If there are multiple TX sources that occupy different slots, you could expect each to have a different bit clock phase (and slightly different frequency too). The transitions may be where one TX is ramping down its power while another ramps up.

Perhaps you can show plots of your waveforms with transition regions.

hello six20star,

First problem is what do you do if there is significant doppler offset due to moving platforms. Most FSK modems use a preamble to be used to detect start of message as well as resolve and remove unknown Doppler. 

The eye diagram differs from what you expect because you have not been observing the output of a matched filter. The difference between successive samples of the complex envelope is a noisy process and does not gather the signal energy to improve SNR.

We have a number of fun options. First, down convert the center frequency between the two FSK tones to zero. Then the upper tone is a positive freq and the lower tone is a negative frequency. You know the center frequencies of the two tones and you know the matched filter for each tone. Build both matched filters and select the tone with the largest output amplitude. A fun trick is to use the derivative of atan equation 

(x * y_dot - y * x_dot)

This avoids atan but still outputs derivative of angle without ever computing angle.

need clean band-width limited derivative filters with odd number of taps to access delayed x and y for time alignment with x_dot and y_dot. We now have a positive and negative rectangle for pos freq and for neg freq tones. can match filter these.   

Sorry for the sidebar but I have a question for fharris (or anyone with experience in this area). I have been analyzing a burst FSK signal that uses a preamble consisting of several phase reversal symbols; that is, a modulated bit pattern of 10101010...

While this preamble design is seemingly good for symbol clock recovery and Doppler offset, to me it appears to be a very poor pattern for detecting start of message because the time domain response of a matched filter for this pattern produces many sidelobes of comparable amplitude. Further I'd say that detecting start of message would be desired *before* acquiring symbol clock or Doppler offset.

My question: Am I missing something or do you think that a modem designed for this signal should use something other than a phase reversals preamble to detect start of message? More broadly what start of message techniques might I consider on this type of burst FSK?

The alternating phases in the preamble is a perfect way to estimate the Doppler and then remove it. The matched filter for the alternating phase sequence then can be easily applied for start of message. See attached. also see MIL_STD_188_181C if you can access it on line... the B extension is easily available.

Continuous Phase Modulation.pdf

fred h