"Computing the mean and standard deviation for a dit frame is very low cost and very easy to implement in a DSP setting on an MCU. Computational cost is one multiply and 2 sums per frame sample plus 3 multiplications, a sum and square root per dit frame.".
Wow, implementing this math is too heavy for my taste. Anyway, if I didn't get it wrong, if after calculating the square root there comes a level comparison, I think you can use the variance you mentioned instead of the standard deviation with no problem. This way you save extracting a square root, the most involved step. Is this correct?
As to Hayward's articles on CCW, there are some of them (thanks to George N2APB) in:
https://midnightdesignsolutions.com/ccw/ccw_news/
As there are more than 120 text files, I downloaded the whole subpage and uploaded it as a .zip to:
/g/qrptech/files/ccw_news.zip
Downloading and decompressing it, one can make a search for "Hayward" or whatever on all the files. I'm afraid, however, that some issues and pictures are missing.
Daniel Perez LW1ECP
24 de marzo de 2024, 11:42:32 ART, Reginald Beardsley via groups.io <pulaskite@...> escribi¨®:
In the CCW articles I recall by Wes Hayward,? (but can't find #$%$) CCW was described as integrating over dit frames and comparing the sums.? However, I recall no mention of how much of a difference in level constituted a signal.
That was where I started a month or so ago at implementing CCW and QRSS.? The problem was I had no way to calculate expected results ready to hand.? So I stepped back and thought about the concept,? "the power level in this dit frame is different from this other dit frame" and how best to measure that in real time at minimum cost.? That led me back to basic physics, probability theory and statistics.? Because of my seismic background my first inclination is to slap things around with some transforms.? But that is not MCU friendly.
Computing the mean and standard deviation for a dit frame is very low cost and very easy to implement in a DSP setting on an MCU.? Computational cost is one multiply and 2 sums per frame sample plus 3 multiplications, a sum and square root per dit frame.? Then make a decision based on measured statistics of the QRN whether to emit an audio tone or not.? It introduces a single dit frame latency.? The output is either a pure side tone signal or silence.? No hiss.
In the absence of a signal, we expect the QRN to be a Gaussian random process in the BW of the channel.? More particularly, we have an expectation of both the mean and the variance.
By initializing the receiver for 10 seconds on frequency to record actual QRN and determine the mean and standard deviation we now have a statistical criterion by which to eliminate QRN.? ? By setting a threshold value for the mean value at a specified standard deviation we can distinguish if a very low amplitude signal is present in addition to the noise.? As the Gaussian is well tabulated one can set the threshold for any desired level of spurious noise as a BER in? percentage probability.
The clock frame timing is the key to making it work.? If the receiving dit frame is off by 50% there will be no change.? Hence the need for synchronization.? That will be a separate topic.
The standard deviation of the dit frame comes in to play in the presence? of QRM.? By tracking? the variance of the variance, one can subject noisy dit frames to a different criterion based on the variance.? If only a constant sinusoidal signal is present, the change in variance will be small.? If a dit frame has a noise burst present, the increase in the variance can be used to select a different threshold level or other processing.
An effective QRM filter for rejecting periodic noise bursts could be developed on top of the basic clocked frame by developing another layer for burst noise based on the variance of the dit frame estimates of the mean and standard deviation among frames.? ? ? ? ? ?
Have Fun!
Reg? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?
Daniel Ricardo Perez LW1ECP