014. Characterize partially observable frequency translating systems

The problem:

To enable 5G+ communication, the level of integration of the building blocks will continue to increase. Grouping more and more functions on a single chip is a massive challenge for characterisation: unobservable internal signals perform signal operations inside the chip that influence the input and output signals. A proper characterisation hence requires estimating and removing the influence these components. In the RF environment where synchronisation of the signals is the main enabler, this represents a major challenge. The presence of unknown signals is a major roadblock for the characterisation of RF detection circuitry such as an IQ detector that is used in complex signal demodulation. Up to now, there is no solution to the problem and the characterisation either relies solely on simulations or on measurements that are plagued with large perturbations.

The idea

To estimate the influence of the unknown signals, an idealized model is used. Next, the deviation between the ideal model and reality is quantified. The characterization then merges the ideal model and its deviation into a truthful representation of the device. For the detector mentioned before, the ideal model is an ideal mixer. The full model is then called a best mixer approximation. Up to now, this model is extracted when all signals are known, but is can be extended to cope with some unknown inputs.

Your contribution:

To characterise RF systems with unknown inputs you will

  • Design, simulate and build a test detector allowing to measure all signals. This will then be used to evaluate the performance of the algorithms operating with partial knowledge only.
  • Implement the best mixer approximation and apply it to your test system
  • Extend the best mixer approximation to handle unobserved internal frequency conversion signals
  • Apply the algorithms to simulation first, and the measurements afterwards. Tune the algorithms whenever necessary and compare to classical methods
  • Finally, design, build and characterize a real IQ detection component where only the input RF signal and the output demodulated signal are observable.


Yves Rolain
Back to top