What's the best method for BPSK demodulation?

[konradb]

I am trying to demodulate a phase shift signal.
The (mixed down) carrier varies from 25KHz to 34KHz. There are actually 4 channels of 3KHz. The signal is a burst of 150ms carrier followed by 500ms of phase data which is +30deg or -30deg.

I sample the data at 500KHz, and store to ram to work on.

There are a few methods that I have tried that work, but I would like expert opinion.

1. Lock a PLL to the pre-amble carrier, then mix the PLL with the signal, then low pass filter.
Not so good results, slight errors in the lock frequency give poor results

2. Take the pre-amble carrier, concatenate it and mix this with the signal, then low pass filter.
Better results but mismatch of sampling for each section also gives less than perfect results

3. Take arctan(Q/I) and unroll, i.e. remove pi/2. Find the slope of the pre-amble, subtract the pre-amble slope from the unrolled data.
Best results

4. Find the FFT of the signal. Shift the carrier to 0Hz. Inverse FFT.
Again, good results

Which method would be recommended, or are there other better option?

 

[alainsan]

Re: BPSK demodulation

Many algorithms will work when the input signal to noise ratio is high.
Not many algorithms will work reliably at low SNRs. So it really depends on the environment for which you are designing this PSK demodulator.

I would do the following:
a) detect the unmodulated preamble with a complex FFT with at least 256 points.
The signal to noise ratio should be good enough for reliable detection of the carrier frequency and phase (peak magnitude).

b) implement a costas loop and pre-position the NCO with the frequency and phase detected by the FFT. The costas loop will track any change in phase during the data portion of the burst, even when the SNR is poor.

Then there is the issue of recovering the symbol timing to sample the PSK signal at the symbol center..... Is there a synchronization pattern in the received signal?

Alain Z.
comblock.com