>>>Hi,
>>>
>>>I have a sinusodal signals with fixed frequencies(multi-tone).
>>>But the signal is randomly on and off.
>>>What I did is that once the signal is off, I discarded the portion
and,
>>>then, when the signal is on, I attached the signal afterward.
>>>So, now, the signal becomes phase discontinued, same frequency,though.
>>>
>>>My frequency tracking algorithm works with the original sinusoidal
>>>signals(without off). Not it cannot track the phase-discontinued
signal
>>>properly.
>>>Is there any way to track the frequency of the signal?
>>>
>>>Thank you in advance.
>>>timothy
>>>
>>
>>Hi,
>>
>>How do you track the frequencies?
>>If you use a fft-based method (looking for peaks),
>>I think it might be better if you leave the "off-parts"
>>in. The multiplies with zeros for the "off-parts"
>>will introduce spurious peaks, but you could
>>use an algorithm which "puts a question-mark" to all new
>>peak in a new STFT frame and just considers to track them
>>if they appear in more than one succesive STFT frame.
>>
>>It really depends on:
>>
>>1. What method are you using to track the frequencies of
>> the sinusoids?
>
>I use DFT filters to track frequency. FFT doesn't work in my case
because
>I need precise frequency tracking.
>
>Furthermore, the off signal may have different frquency, i.e on freq is
>15.65 Mhz and off freq is 16.59 Mhz. In this case, I have to track 15.65
>Mhz freq. of course it has on/off switches.
>
>>
>>2. What is the ratio between "on" and "off" parts?
>> How much of the signal is dropped?
>
>It is RANDOM for now.

Ok, so you want to track frequencies and be independent of
sudden phase discontinuities.
Another factor is the spacing of the frequencies in your input signal.
Have you considered to split the signal into bands with complex
values (with complex coefficient bandpass filters - like a hilbert
transform but with narrow passband) and track the frequencies
by averaging delta phase values.
That would just work if the frequencies are far enough away from
each other. That idea would be based on isolating frequency regions
containing sinusoids to track, so that you would have _one_ sinusoid
to track per region. In that case averaged instantanious frequencies
would be usefull and the distortion due to the drop-outs have very
small extend in time (since the instantanious freq of a single sinusoid
can be determined from just two complex valued output samples)
This would be very sensitive to noise (that's why I say _averaged_
instantanious freqs) and it's probably closely related to what you are
doing now. I'm just trying to suggest another way to look at
it (the goertzel/dft filter thing. vs. narrow complex bandpass filters
followed by delta phase measurement).

gr.
Bjoern