In optical coherence tomography (OCT), unbiased and low variance Doppler frequency estimators are desirable for
blood velocity estimation. Hardware improvements in OCT mean that ever higher acquisition rates are possible.
However, it is known that the Kasai autocorrelation estimator, unexpectedly, performs worse as acquisition rates
increase. Here we suggest that maximum likelihood estimators (MLEs) that utilize prior knowledge of noise
statistics can perform better. We show that the additive white Gaussian noise maximum likelihood estimator
(AWGN MLE) has a superior performance to the Kasai autocorrelation estimate under additive shot noise
conditions. It can achieve the Cramer-Rao Lower Bound (CRLB) for moderate data lengths and signal-to-noise
ratios (SNRs). However, being a parametric estimator, it has the disadvantages of sensitivity to outliers, signal
contamination and deviations from noise model assumptions. We show that under multiplicative decorrelation
noise conditions, the AWGN MLE performance deteriorates, while the Kasai estimator still gives reasonable
estimates. Hence, we further develop a multiplicative noise MLE for use under multiplicative noise dominant
conditions. According to simulations, this estimator is superior to both the AWGN MLE and the Kasai estimator
under these conditions, but requires knowledge of the decorrelation statistics. It also requires more computation.
For actual data, the decorrelation MLE appears to perform adequately without parameter optimization. Hence
we conclude that it is preferable to use a maximum likelihood approach in OCT Doppler frequency estimation
when noise statistics are known or can be accurately estimated.