Warren C. Pratt, NR0V: Advanced Algorithms for Noise Blanking and Noise Reduction
Wideband noise-blankers and LMS noise-reduction algorithms are commonplace in modern SDRs. With today's CPU power, more advanced algorithms offering superior performance are also available. Warren focuses on two such algorithms implemented in 2015 in the WDSP library used for the openHPSDR program: 1. The Spectral Noise Blanker uses linear predictive coding and often removes impulse noise under conditions where wideband blankers are ineffective. Impulses are detected by comparing the observed waveform with a predicted waveform. Impulses are corrected by recreating an estimate of corrupt portions of the original waveform using spectral information. 2. The Spectral Noise Reduction algorithm operates in the frequency domain and, based upon statistical models of speech and noise, reduces random noise much more effectively than LMS algorithms. The seminal work for this approach was published by Yariv Ephraim and David Malah in 1984. However, the state of the art has advanced substantially over the past thirty years.