This is a quick demonstration of an experimental voice squelch algorithm for Thetis SDR software. The software was developed by Dr. Warren Pratt, NR0V. I helped a little by analyzing a hardware version of the algorithm in order to provide data for Warren's efforts.
The intent of the voice squelch algorithm is to provide further improvement to the listening experience over and above the already powerful NR2 noise reduction algorithm. NR2 was also developed by Warren and it continues to be the only example of spectral estimation noise reduction to be found anywhere in amateur radio.
The squelch algorithm discriminates human speech from noise in a two step process. First the algorithm does a band-limited frequency-to-voltage (F-to-V) conversion. The resulting voltage varies according to the primary frequency content of the input signal. The time varying output of the F-to-V conversion is then subjected to a peak-to-average ratio (PAPR) measurement. When the PAPR exceeds an adjustable threshold the squelch is unmuted. There are also the usual squelch-related time constants involved in determining how long the PAPR threshold must remain exceeded before unmute, and how long it must remain below threshold before muting.
With the use of NR2 noise reduction and proper adjustment of AGC gain, and on a band with a stable noise floor, signals with as little as 6dB (1 S unit) of SNR can by reliably processed by the voice squelch. The resulting elimination of noise, even between words on an SSB voice transmission, can substantially reduce listening fatigue. The excellent sensitivity of the voice squelch algorithm allows for the detection of all but the smallest breaking stations when in a "rag chew" type QSO.
This algorithm can't overcome static crashes or an unstable noise floor, nor can it filter out ionosondes and other interfering signals. But on a band with a stable noise floor it works very, very well.
22 апр 2023
