With Vaishnavi Adella (now at Qualcomm, India), Sai Charan Thoutam (now at Qualcomm, India)
Implemented the paper : Biased estimators with adaptive shrinkage targets for orthogonal frequency division multiple access channel estimation
- Channel Estimation is the means of characterising channel effects-scattering, fading and power decay
- Can be done using either decision feedback scheme or the known pilot symbols
- Here : Using user specific pilots
- Why is it crucial ? Huge transmit power spent on it Incorrect estimates lead to residual cancellation errors Necessary for high data rates
- Channel statistics – known
- Estimation methods such as modified least squares require wide band pilots
- Finite or large number of pilots / wide band pilots
- The two-dimensional (2D)-minimum mean square error (2D-MMSE) methods [1, 2] can be applied using the pilots in the RB. However, optimal MMSE estimator requires the knowledge of the channel statistics which are seldom known accurately at the receiver.
- MVUE-Zero bias as constraint and acheive CRLB
- ML estimate is biased in order to reduce the MSE.
- 2D-MMSE can be used when channel statistics is unknown
- Assumption: Ideally band limited and time limited uniform scattering function Is that even possible ? Google says 'the only time and band limited signal is zero'. A little more of research and we find: A time limited signal with most of its energy contained in the band is a good approximation for both time and band limited signal.
- 2D-MMSE : Performance degrades if the robust filter has finite number of taps [2]
- JS-estimator used to bridge the gap between robust 2D MMSE and optimal MMSE
- Applying biased estimation techniques for localised CFR estimation
- Adaptively choosing a vector shrinkage target for the biased estimator that best reflects the time–frequency selectivity of the CFR using the aid of multiple hypothesis tests
- Choosing the thresholds for the hypothesis tests such that the probability of incorrect choice of shrinkage targets is bounded.