You are not logged in.
Since Ive been having issues getting the OFDM Reference Design working in multiplex mode I decided to go back to warplab and look at some of the constellation of those simple examples. I picked the siso_example_TxRx first and at the end took the Txdata and Rxdata and used scatterplot to see the constellation. The Tx constellation was what I expected, several tightly compact symbols spaced evenly in a circle around the origin. The Rxdata constellation was however more surprising, the symbols were smeared into one big ring around the origin, no distinction between one symbol to the next. The same happened when I tried the mimo_example_TxRx. I asked my advisor and he said a timing issue usually causes something like this.
So has this been seen before? Did I maybe use scatterplot incorrectly? Did I maybe plot the wrong data? Or is there really some timing problem between the Tx warp and the Rx warp?
To be clear this is what I added at the end of the warplab examples: "figure; scatterplot(NodeX_RadioY_RxData);"
Offline
You're seeing the effects of carrier frequency offset. Each node uses a local oscillator to generate its RF carrier, and each operates with slightly different frequencies. The result is that the baseband signal output from the receive radio is shifted slightly above/below 0Hz. This is a standard "feature" (really, an unavoidable complication) of wireless systems.
For the WARPLab comm examples, differential encoding is used to combat the CFO. This works because the magnitude of the CFO is much smaller than the symbol rate, so phase differences between consecutive symbols can still be used to encode information.
In the OFDM reference design, the Rx PHY estimates the CFO and corrects it before taking the FFT. It also tracks residual phase offsets using pilot tones embedded in each OFDM symbol (same techniques as in 802.11a OFDM).
Offline
Thanks Murphpo, that makes more sense. So in the future if I want to demonstrate the constellation of a warplab example how might I go about it?
Offline
You would have to implement some kind of CFO estimation/compensation scheme. The details depend entirely on the PHY design. For the WARPLab comm examples, you could use a data-driven scheme like a Costas Loop. You could also modify the examples to estimate the phase/frequency offsets using the known values of the first few symbols, then apply a correction to the rest. This is analogous to the real-time PHY's use of a common preamble to estimate CFO.
Offline