New SAEM algorithm cleans radio telescope images 2x better than Gaussian methods
Robust state-space imaging handles heavy-tailed RFI noise without losing fidelity...
Deep Dive
A new method uses a Stochastic Approximation Expectation-Maximization (SAEM) algorithm for robust state-space radio interferometric imaging, handling heavy-tailed compound-Gaussian noise via closed-form Gibbs sampling. Numerical experiments show it significantly improves reconstruction fidelity and robustness to radio-frequency interference, outperforming a Gaussian EM algorithm and even an oracle RTS smoother.
Key Points
- SAEM algorithm uses closed-form Gibbs sampling to handle compound-Gaussian noise common in RFI-affected radio data
- Outperforms Gaussian EM and even an oracle RTS smoother in reconstruction fidelity under heavy-tailed interference
- Requires no extra calibration—works directly with existing interferometric array data
Why It Matters
More robust radio astronomy imaging means clearer views of faint cosmic sources despite growing satellite interference.