I think we don't have to do carrier phase recovery. Do i need to do carrier frequency recovery. My symbol rate is high around 2 MHz so i am thinking carrier freq offset is less compared the symbol rate. Or does my gardner timing recovery can take care of this carrier freq offset recovery. I have another question should i do differential decoding then timing recovery or timing recovery followed by differential decoding . Because the pi/4 dqps will have valid symbols at zero level also ie symbol with zero I data and Q non zero data. Similary zero Q data and I non zero data. So i am thinking if do timing recovery first i may get proper output as garder looks for zeros crossing between symbols. Is my explanation right

  • $\begingroup$ It would be good if you could clarify the application or what is most important to you, cost, power, size etc. as that drives the architecture. $\endgroup$ – Dan Boschen Dec 18 '19 at 3:30

Yes you are correct, differential modulations such as DBPSK, DQPSK, $pi/4$ DQPSK do not need to perform carrier recovery as you can compare one symbol to the next to perform demodulation using a delay and multiply approach. This results in a simple and potentially low power receiver, at the expense of performance compared to coherent modulation schemes.

As far as a static carrier offset this should be no issue- the worst case is if the carrier frequency is changing from one symbol to the next (and at some level it always is, in particular phase noise)- in this regard simply consider how much your symbol rotates from one symbol to the next when you demodulate with this approach and you can directly see the impact of frequency offset (and if the SNR degradation from that rotation based on the resulting symbol distance would be acceptable). For example, a 2 degree rotation would be negligible, while with a 45 degree rotation you be on the threshold of an incorrect symbol decision with no other added noise (using the demodulation approach I describe below). The impact of this can be quantified from the phase noise, and how I approach that with a differential demodulation is to consider with regards to phase how a multiplier for small angles is a "phase adder" thus the delay and multiply demod process is a high pass filter with regards to phase noise. (The filter is given by $1-z^{-1}$). It has a gain of 2 at the high frequency portion (symbol rate) and a null at DC- so it passes (and doubles) the phase noise at those higher offsets). Ultimately you can integrate the high pass filtered phase noise from DC out to the symbol rate and double that to determine the rms phase error, and determine performance issues with consideration to the impact of phase offsets as I described above. What you will find in doing this is the demodulator will likely be very robust against phase noise.

An approach I would consider is doing a complex conjugate multiplication of each new symbol with the previous symbol with the additional +/-45 degree rotation in that process to remove the pi/4 offsets. The output of this would be I and Q with bi-phase signaling on each (similar to QPSK) that you could easily pass into a Gardner (at 2 samples per symbol) for timing recovery followed by matched filters and then data decision. An alternate approach used for low cost / low power non-coherent pi/4 QSPK demodulation consists of hard limiting the waveform and then using a simple FM discriminator followed by a four level detector (since the 4 phase transitions +45°, -45°, +135°, -135° from each symbol to the next correspond to 4 possible frequencies).

| improve this answer | |
  • $\begingroup$ Hi Dan , this is exactly what I was thinking. So if i have 4 samples/symbol i probably have to do complex multiplication and 45 degrees rotatiion for each of the samples separately .so it would be downsample process and upsample and then passvto Ted?? . So this is the way it is normally done. So timing recovery should not be done before diff. Decoding?? $\endgroup$ – Jeremy Dec 18 '19 at 3:44
  • $\begingroup$ If you have sufficiently filtered you signal you should be able to do it all at 2 samples per symbol (that really is a matter of filter complexity and that you have achieved sufficient out of band rejection-- if not filter, decimate to 2 samples per symbol and proceed from there with the rest of your receiver). $\endgroup$ – Dan Boschen Dec 18 '19 at 3:48
  • $\begingroup$ As far as the differential decoding I am not confident to answer that at the top of my head without tracing the whole transmit to receiver path again; starting with the differential encoding, the differential demodulation as I described using the full complex conjugate delay and multiply and to what extent a further differential decoding is needed to eliminate symbol ambiguity---but if you trace that out it should be clear. However that said I don't see how it impacts that timing recovery would be immediately done on the I and Q samples at the output of the multiplier 2 samples/symbol using... $\endgroup$ – Dan Boschen Dec 18 '19 at 3:51
  • $\begingroup$ ...TED followed by matched filter followed by decisions--- subsequent differential decoding to the extent needed would be done on the decisions. $\endgroup$ – Dan Boschen Dec 18 '19 at 3:52
  • $\begingroup$ If you did the differential demodulation as I describe- what carrier frequency would that be at? That alone may be one reason you have a higher sample per symbol rate until you get to the multiplier output. $\endgroup$ – Dan Boschen Dec 18 '19 at 3:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.