2

What actual protocols exist out there that employ Offset Quadrature Phase-Shift Keying (OQPSK)?

What do they use for the frame synchronization marker (or "preamble" or "sync word" or "flag")?

Do they use a unique word, a unique pattern of bits? If yes, do they do some kinda bogus bit insertion (similar to the zero-insertion what the old SDLC protocol did)? If not, how can they guarantee that the data never mimics the frame sync marker and a spurious frame sync occurs?

What do they do for Offset QPSK to synchronize the I/Q phase of the intermediate frequency (IF) signal? Does the frame sync marker give the receiver an unambiguous signal from which to derive the correct I/Q reference phase?

Lotsa questions. I've been thinking of OQPSK a bit and have rambled out loud about it at the DSP Stack Exchange. OQPSK naturally pairs the bits from a single serial stream into "dibits". The even and odd bits naturally are routed to the I and Q signals without any contrived pairing of bits. Then it seems to me that the 16 bits 0011001100110011 would be the natural marker for the lowest instantaneous frequency and 0110011001100110 would be the highest frequency. One pattern could be the frame start marker and the other pattern a channel idle marker.

But then the 15 bits 011001100110011 should never occur in the data and a bogus 0 should be inserted after the 14 bits 01100110011001 so that neither marker would ever appear in the data. In the style of SDLC, the receiver would always remove that bogus 0 that occurs after 01100110011001. If, before the zero-removal, a 1 appears after 01100110011001, then I think either a channel idle or frame start marker is occurring (one more bit is needed on one side or the other of the 15 bits 011001100110011 to know for sure).

But this paper suggests other bit patterns (Barker sequences) for markers.

What are the protocols in use? What is the wisdom of those here that have actually worked on OQPSK at a low level?

robert bristow-johnson
  • 1,645
  • 1
  • 12
  • 29
  • 1
    The latest work in OQPSK has been SOQPSK (-TG & -MIL & probably others) due to its excellent spectral efficiency for military applications. There are a good number of recent papers out there on SOQPSK, maybe they can answer your questions. BTW, barker codes have really nice correlation properties: namely a nice unambiguous peak. Using a correlator on the complex I/Q samples, at the correlation peak you can pluck off the carrier phase and kickstart carrier phase correction and tracking. – Andy Walls Oct 09 '18 at 03:03
  • @AndyWalls, what's the "**S**" stand for? – robert bristow-johnson Oct 09 '18 at 04:54
  • okay, "Shaped". shaped what? pulse? spectrum? still haven't figgered out what the "**TG**" is. – robert bristow-johnson Oct 09 '18 at 04:59
  • telemetry group. – robert bristow-johnson Oct 09 '18 at 05:06
  • [this masters thesis](https://kuscholarworks.ku.edu/bitstream/handle/1808/7644/ReaZanabria_ku_0099M_11339_DATA_1.pdf) is helpful. so it's 8 points in the constellation instead of 4. so it's a little less noise immune (for the same transmit power) but moves more data in the same bandwidth than does OQPSK? – robert bristow-johnson Oct 09 '18 at 05:15
  • it's interesting @AndyWalls that barker codes are considered good for QPSK. i can see the preferable autocorrelation quality for BPSK but for QPSK or OQPSK the neighboring bits are not correlated against each other in the matched filter that would be used to detect a frame sync marker. – robert bristow-johnson Oct 11 '18 at 07:28
  • You can have a correlator (aka matched filter) in front of a split into I & Q branches for preamble detection and carrier phase estimation. (I do that with MSK which is equivalent to an OQPSK.) But to do correlation after a split into I & Q, you are right, you would have to use a point in the processing chain after the bits from I & Q are reinterleaved. – Andy Walls Oct 11 '18 at 10:56
  • well, correlation on the **bits** is really nothing other than a bit-wise exclusive-OR on the word. you look for a match or not. i don't see where Barker codes help that. the issue for me is a marker that somehow lives at the extreme (so other "more normal" data will not be at that extreme) has easily-recognized analog properties in the modulated form, and can be used to sync up the IF of what we used to call the "local oscillator". – robert bristow-johnson Oct 11 '18 at 20:09
  • the other thing is that i cannot see how OQPSK is any less bandwidth efficient than anything else. each even bit goes to \$i(t)\$, each odd bit goes to \$q(t)\$ and each bit is mapped to \$\{-1,+1\}\$ and that is mapped to a \$\mathrm{sinc}()\$ function. if the data bits are well scrambled, that occupies the entire bandwidth. so i still haven't figgered out the advantages of SOQPSK over OQPSK except there are 8 points in the constellation, that gives you one more bit per symbol period, but at the expense of noise margin. – robert bristow-johnson Oct 11 '18 at 20:15
  • so @AndyWalls, if the correlation is done *only* to the re-interleaved stream, how is the **phase** of the combined IQ signal derived from the received signal? i presume this is a **real** signal bumped up into the intermediate frequencies some where. – robert bristow-johnson Oct 11 '18 at 20:20

0 Answers0